chapter 10 artificial intelligence. 2 chapter 10: artificial intelligence 10.1 intelligence and...
Post on 22-Dec-2015
223 views
TRANSCRIPT
Chapter 10
Artificial Intelligence
2
Chapter 10 Artificial Intelligence
101 Intelligence and Machines 102 Understanding Images 103 Reasoning 104 Artificial Neural Networks 105 Genetic Algorithms 106 Other Areas of Research 107 Considering the Consequences
3
Intelligent agents
Agent = ldquodevicerdquo that responds to stimuli from its environment Sensorsmdashreceive data from environment --microphone camera Actuatorsmdashresponse to the stimuli --legs wings hands
The goal of artificial intelligence is to build agents that behave intelligently
4
Levels of intelligence in behavior
Reflex actions are predetermined responses to the input data
Intelligent response actions affected by knowledge of the environment
Goal seeking Learning
5
Artificial intelligence research approaches
Agentrsquos intelligence is judged by observing its input-response patterns Performance oriented Researcher
tries to maximize the performance of the agents
Simulation oriented Researcher tries to understand how the agents produce responses
6
Turing test
Proposed by Alan Turing in 1950 Benchmark for progress in artificial
intelligence Test setup Human interrogator
communicates with test subject by typewriter
Test Can the human interrogator distinguish whether the test subject is human or machine
7
Figure 101 The eight-puzzle in its solved configuration
8
Figure 102 Our puzzle-solving machine
9
Techniques for understanding images
The first intelligent behavior required by the puzzle-solving machine is the extraction of information through a visual medium1048708
Unlike photographing an image the problem is to understand the image (Computer Vision) ndashthe ability to perceive1048708
Since the possible images are finite the machine can merely compare the different sections of the picture to prerecorded templates pixel by pixel and reveal the condition of the puzzle1048708
10
Techniques for understanding images
Optical readers apply the similar method for image recognition (hand-writing)1048708
A certain degree of uniformity(style size orientation non-overlapping hellip) is required1048708
The alternative is to first extract the geometric features(digit 1 a single vertical line) and make comparison in terms of these features
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
2
Chapter 10 Artificial Intelligence
101 Intelligence and Machines 102 Understanding Images 103 Reasoning 104 Artificial Neural Networks 105 Genetic Algorithms 106 Other Areas of Research 107 Considering the Consequences
3
Intelligent agents
Agent = ldquodevicerdquo that responds to stimuli from its environment Sensorsmdashreceive data from environment --microphone camera Actuatorsmdashresponse to the stimuli --legs wings hands
The goal of artificial intelligence is to build agents that behave intelligently
4
Levels of intelligence in behavior
Reflex actions are predetermined responses to the input data
Intelligent response actions affected by knowledge of the environment
Goal seeking Learning
5
Artificial intelligence research approaches
Agentrsquos intelligence is judged by observing its input-response patterns Performance oriented Researcher
tries to maximize the performance of the agents
Simulation oriented Researcher tries to understand how the agents produce responses
6
Turing test
Proposed by Alan Turing in 1950 Benchmark for progress in artificial
intelligence Test setup Human interrogator
communicates with test subject by typewriter
Test Can the human interrogator distinguish whether the test subject is human or machine
7
Figure 101 The eight-puzzle in its solved configuration
8
Figure 102 Our puzzle-solving machine
9
Techniques for understanding images
The first intelligent behavior required by the puzzle-solving machine is the extraction of information through a visual medium1048708
Unlike photographing an image the problem is to understand the image (Computer Vision) ndashthe ability to perceive1048708
Since the possible images are finite the machine can merely compare the different sections of the picture to prerecorded templates pixel by pixel and reveal the condition of the puzzle1048708
10
Techniques for understanding images
Optical readers apply the similar method for image recognition (hand-writing)1048708
A certain degree of uniformity(style size orientation non-overlapping hellip) is required1048708
The alternative is to first extract the geometric features(digit 1 a single vertical line) and make comparison in terms of these features
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
3
Intelligent agents
Agent = ldquodevicerdquo that responds to stimuli from its environment Sensorsmdashreceive data from environment --microphone camera Actuatorsmdashresponse to the stimuli --legs wings hands
The goal of artificial intelligence is to build agents that behave intelligently
4
Levels of intelligence in behavior
Reflex actions are predetermined responses to the input data
Intelligent response actions affected by knowledge of the environment
Goal seeking Learning
5
Artificial intelligence research approaches
Agentrsquos intelligence is judged by observing its input-response patterns Performance oriented Researcher
tries to maximize the performance of the agents
Simulation oriented Researcher tries to understand how the agents produce responses
6
Turing test
Proposed by Alan Turing in 1950 Benchmark for progress in artificial
intelligence Test setup Human interrogator
communicates with test subject by typewriter
Test Can the human interrogator distinguish whether the test subject is human or machine
7
Figure 101 The eight-puzzle in its solved configuration
8
Figure 102 Our puzzle-solving machine
9
Techniques for understanding images
The first intelligent behavior required by the puzzle-solving machine is the extraction of information through a visual medium1048708
Unlike photographing an image the problem is to understand the image (Computer Vision) ndashthe ability to perceive1048708
Since the possible images are finite the machine can merely compare the different sections of the picture to prerecorded templates pixel by pixel and reveal the condition of the puzzle1048708
10
Techniques for understanding images
Optical readers apply the similar method for image recognition (hand-writing)1048708
A certain degree of uniformity(style size orientation non-overlapping hellip) is required1048708
The alternative is to first extract the geometric features(digit 1 a single vertical line) and make comparison in terms of these features
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
4
Levels of intelligence in behavior
Reflex actions are predetermined responses to the input data
Intelligent response actions affected by knowledge of the environment
Goal seeking Learning
5
Artificial intelligence research approaches
Agentrsquos intelligence is judged by observing its input-response patterns Performance oriented Researcher
tries to maximize the performance of the agents
Simulation oriented Researcher tries to understand how the agents produce responses
6
Turing test
Proposed by Alan Turing in 1950 Benchmark for progress in artificial
intelligence Test setup Human interrogator
communicates with test subject by typewriter
Test Can the human interrogator distinguish whether the test subject is human or machine
7
Figure 101 The eight-puzzle in its solved configuration
8
Figure 102 Our puzzle-solving machine
9
Techniques for understanding images
The first intelligent behavior required by the puzzle-solving machine is the extraction of information through a visual medium1048708
Unlike photographing an image the problem is to understand the image (Computer Vision) ndashthe ability to perceive1048708
Since the possible images are finite the machine can merely compare the different sections of the picture to prerecorded templates pixel by pixel and reveal the condition of the puzzle1048708
10
Techniques for understanding images
Optical readers apply the similar method for image recognition (hand-writing)1048708
A certain degree of uniformity(style size orientation non-overlapping hellip) is required1048708
The alternative is to first extract the geometric features(digit 1 a single vertical line) and make comparison in terms of these features
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
5
Artificial intelligence research approaches
Agentrsquos intelligence is judged by observing its input-response patterns Performance oriented Researcher
tries to maximize the performance of the agents
Simulation oriented Researcher tries to understand how the agents produce responses
6
Turing test
Proposed by Alan Turing in 1950 Benchmark for progress in artificial
intelligence Test setup Human interrogator
communicates with test subject by typewriter
Test Can the human interrogator distinguish whether the test subject is human or machine
7
Figure 101 The eight-puzzle in its solved configuration
8
Figure 102 Our puzzle-solving machine
9
Techniques for understanding images
The first intelligent behavior required by the puzzle-solving machine is the extraction of information through a visual medium1048708
Unlike photographing an image the problem is to understand the image (Computer Vision) ndashthe ability to perceive1048708
Since the possible images are finite the machine can merely compare the different sections of the picture to prerecorded templates pixel by pixel and reveal the condition of the puzzle1048708
10
Techniques for understanding images
Optical readers apply the similar method for image recognition (hand-writing)1048708
A certain degree of uniformity(style size orientation non-overlapping hellip) is required1048708
The alternative is to first extract the geometric features(digit 1 a single vertical line) and make comparison in terms of these features
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
6
Turing test
Proposed by Alan Turing in 1950 Benchmark for progress in artificial
intelligence Test setup Human interrogator
communicates with test subject by typewriter
Test Can the human interrogator distinguish whether the test subject is human or machine
7
Figure 101 The eight-puzzle in its solved configuration
8
Figure 102 Our puzzle-solving machine
9
Techniques for understanding images
The first intelligent behavior required by the puzzle-solving machine is the extraction of information through a visual medium1048708
Unlike photographing an image the problem is to understand the image (Computer Vision) ndashthe ability to perceive1048708
Since the possible images are finite the machine can merely compare the different sections of the picture to prerecorded templates pixel by pixel and reveal the condition of the puzzle1048708
10
Techniques for understanding images
Optical readers apply the similar method for image recognition (hand-writing)1048708
A certain degree of uniformity(style size orientation non-overlapping hellip) is required1048708
The alternative is to first extract the geometric features(digit 1 a single vertical line) and make comparison in terms of these features
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
7
Figure 101 The eight-puzzle in its solved configuration
8
Figure 102 Our puzzle-solving machine
9
Techniques for understanding images
The first intelligent behavior required by the puzzle-solving machine is the extraction of information through a visual medium1048708
Unlike photographing an image the problem is to understand the image (Computer Vision) ndashthe ability to perceive1048708
Since the possible images are finite the machine can merely compare the different sections of the picture to prerecorded templates pixel by pixel and reveal the condition of the puzzle1048708
10
Techniques for understanding images
Optical readers apply the similar method for image recognition (hand-writing)1048708
A certain degree of uniformity(style size orientation non-overlapping hellip) is required1048708
The alternative is to first extract the geometric features(digit 1 a single vertical line) and make comparison in terms of these features
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
8
Figure 102 Our puzzle-solving machine
9
Techniques for understanding images
The first intelligent behavior required by the puzzle-solving machine is the extraction of information through a visual medium1048708
Unlike photographing an image the problem is to understand the image (Computer Vision) ndashthe ability to perceive1048708
Since the possible images are finite the machine can merely compare the different sections of the picture to prerecorded templates pixel by pixel and reveal the condition of the puzzle1048708
10
Techniques for understanding images
Optical readers apply the similar method for image recognition (hand-writing)1048708
A certain degree of uniformity(style size orientation non-overlapping hellip) is required1048708
The alternative is to first extract the geometric features(digit 1 a single vertical line) and make comparison in terms of these features
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
9
Techniques for understanding images
The first intelligent behavior required by the puzzle-solving machine is the extraction of information through a visual medium1048708
Unlike photographing an image the problem is to understand the image (Computer Vision) ndashthe ability to perceive1048708
Since the possible images are finite the machine can merely compare the different sections of the picture to prerecorded templates pixel by pixel and reveal the condition of the puzzle1048708
10
Techniques for understanding images
Optical readers apply the similar method for image recognition (hand-writing)1048708
A certain degree of uniformity(style size orientation non-overlapping hellip) is required1048708
The alternative is to first extract the geometric features(digit 1 a single vertical line) and make comparison in terms of these features
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
10
Techniques for understanding images
Optical readers apply the similar method for image recognition (hand-writing)1048708
A certain degree of uniformity(style size orientation non-overlapping hellip) is required1048708
The alternative is to first extract the geometric features(digit 1 a single vertical line) and make comparison in terms of these features
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
11
Techniques for understanding images
Template matching Image processing identify the
characteristics of the image1048708 Edge enhancement to clarify the
boundary (taking a derivative) Region(with common properties color
hellip) finding for identifying objects Smoothing(removing flawsnoises in
image)1048708
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
12
Techniques for understanding images
Image analysis identify the meaning of these characteristics1048708
-- It is to recognize partially obstructed objects from different perspectives1048708
-- First assumption of what the image might be is made (clue)1048708
-- Then associate the image components with the objects conjectured to exist
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
13
Reasoning
After deciphering the positions of tiles from visual image the remaining task is to move the tiles to reach the final state from the current state1048708
The eight-puzzle has many configurations such that explicitly hard-coded each case for problem solving is not generally feasible
Some algorithm is necessary to resolve the problem in a systematic way1048708
The machine will then ably make decisions draw conclusions and perform elementary reasoning activities
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
14
Components of production systemsA production system classifies the common
characteristics shared by a class of reasoning problems and has the following components
1 Collection of states Start or initial state Goal state
2 Collection of productions rules or moves Each production may have preconditions
3 Control system decides which production to apply next
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
15
Data processing for production systems
State graph = states productions and preconditions
A graph consists of nodes and arcs(arrows) connecting nodes1048708A state graph has nodes representing states and arrows representing rules1048708
The arc linking two nodes signifies two states can be shift to each other using the rule the absence of arcs implicitly indicates the preconditions are not met1048708The problem magnitude may be too large for explicitly showing the entire state graph1048708
Partial representation of the state graph will help understand the problem
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
16
Data processing for production systems
State graph = states productions and preconditions
Search tree = record of state transitions explored while searching for a goal state Breadth-first search Depth-first search
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
17
Figure 103 A small portion of the eight-puzzlersquos state graph
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
18
Figure 104 Deductive reasoning in the context of a production system
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
19
Figure 105 An unsolved eight-puzzle
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
20
Figure 106 A sample search tree
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
21
Figure 107 Productions stacked for later execution
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
22
Figure 108 An unsolved eight-puzzle
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
23
Heuristic strategies
Generally a search tree may grow much huger if the nodesrsquo fan-outs are large 1048708
It becomes more complicate if the goal is very far away (more generations)1048708
Developing a full (exhaustive) search tree[brute-force methods] may be impractical1048708
In contrast to this breadth-first approach (layer by layer) we (humans) may attempt to pursue the more promising paths to greater depths in a depth-first manner (vertical)1048708
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
24
Heuristic strategies
Heuristic strategy is to develop a heuristicndasha quantitative measure on how close a state is to the goal
Requirements for good heuristics Must be much easier to compute than a
complete solution Must provide a reasonable estimate of
proximity to a goal
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
25
Figure 109 An algorithm for a control system using heuristics
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
26
Figure 1010 The beginnings of our heuristic search
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
27
Figure 1011 The search tree after two passes
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
28
Figure 1012 The search tree after three passes
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
29
Figure 1013 The complete search tree formed by our heuristic system
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
30
Neural networks
CPU is not capable of perceive and reasoning
Artificial neuron Each input is multiplied by a weighting
factor Output is 1 if sum of weighted inputs
exceeds a threshold value 0 otherwise Network is programmed by adjusting
weights using feedback from examples
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
31
Figure 1014 A neuron in a living biological system
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
32
Neural networks
ANN are multi-processing architectures to model networks of concurrent neurons1048708Each processing unit in ANN is a simple device to simulate the neuron1048708
The output of the unit may be 0 or 1 (or the fractional numbers in-between) dependent on the whether its effective input exceeds a given threshold value
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
33
Figure 1015 The activities within a processing unit
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
34
Figure 1016 Representation of a processing unit
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
35
Figure 1017 A neural network with two different programs (a) o=1 when 2 inputs diff(b)o=1 when both I=1
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
36
Neural networks
Different weights determine different output values1048708Case (a) will produce 1 if its two inputs differ while (b) outputs 1 if both inputs are 1rsquos1048708
A human brain contains roughly 10^11 neurons with about 10^4 synapses per neuron
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
37
character recognition
A specific application ndashcharacter recognition distinguish C and T regardless of the orientation1048708The network produces a 0 if the recognized letter is C or a 1 if the letter is a T
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
38
Figure 1018 Uppercase C and uppercase T
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
39
Figure 1019 Various orientations of the letters C and T
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
40
Neural networks
The system contains two levels of units1048708The first level has many units one for each 3x3 block of pixels1048708
Each unit has nine inputs the inputs of adjacent units overlap1048708Threshold = 5 centerrsquos weight = 2 othersrsquoweights = -11048708
The second level only has one unit with a separate input for each unit in the first level1048708Threshold = 5 each weight = 11048708
It outputs a 1 iff at least one input is a 1
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
41
Neural networks
If ldquoCrdquois present all the first level units will produce a 01048708All the possible cases can be enumerated
If ldquoTrdquois present only the first levelrsquos unit (highlighted below) will output a 1 while others output 0rsquos1048708The final output is 1
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
42
Figure 1020 The structure of the character recognition system
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
43
Figure 1021 The letter C in the field of view
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
44
Figure 1022 The letter T in the field of view
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
45
Associative memory
Associative memory = the retrieval of information relevant to the information at hand
One direction of research seeks to build associative memory using neural networks that when given a partial pattern transition themselves to a completed pattern
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
46
Figure 1023 An artificial neural network implementing an associative memory
1 The lines connecting circles are two-way connectionie output of one unit is connectedAs input of other unit2 The number associated with Lines are weights3 The number inside the circle is threshold
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
47
Figure 1024 The steps leading to a stable configuration
Two stable states1 Perimeter stable state (later stable state)When we initialize the Network with a least four Adjacent units on the Perimeter in their excitedstates
-11
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
48
The steps leading to a stable configuration
Two stable states2 Center stable state (former stable state)When we initialize the Network with center excited And no more than two of Perimeter in their excitedstates
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
49
Genetic algorithms
Simulate genetic processes to evolve algorithms Start with an initial population of ldquopartial
solutionsrdquo Graft together parts of the best performers
to form a new population Periodically make slight modifications to
some members of the current population Repeat until a satisfactory solution is
obtained
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
50
Figure 1025 Crossing two poker-playing strategies
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
51
Figure 1026 Coding the topology of an artificial neural network
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
52
Language processing
Syntactic analysis(subjectverb noun) Semantic analysis(identify actions) Contextual analysis(understanding)--The bat flew from his handEntire database Information retrieval(web searching) Information extraction(template)
Semantic net(a large linked data structure)
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
53
Figure 1027 A semantic net
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
54
Robotics
Began as a field within mechanical and electrical engineering
Today encompasses a much wider range of activities Robot cup competition Evolutionary robotics
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
55
Expert systems
Expert system = software package to assist humans in situations where expert knowledge is required Example medical diagnosis Often similar to a production system Blackboard model several problem-
solving systems share a common data area
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans
56
Some issues raised by artificial intelligence
When should a computerrsquos decision be trusted over a humanrsquos
If a computer can do a job better than a human when should a human do the job anyway
What would be the social impact if computer ldquointelligencerdquo surpasses that of many humans