artificial neural networks an overview and analysis
TRANSCRIPT
![Page 1: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/1.jpg)
Artificial Neural NetworksArtificial Neural Networks
An Overview and Analysis
![Page 2: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/2.jpg)
Modeling the Human BrainModeling the Human Brain
Artificial Neural Networks concentrate on imitating humans rather than acting as rational agents.
The goal in ANNs is to imitate the learning
that takes place in the brain.
The motivation behind this was to find a way
for a machine to LEARN.
![Page 3: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/3.jpg)
Structure of an ANNStructure of an ANN The basic function component of an ANN is the
unit.
These units are connected to each other through links.
Each link has a weight associated with it. These weights are the means of long term storage in ANNs. Also, learning is usually accomplished by updating these weights.
Some units are connected to the outside world, and are designated as input or output units.
Units on the same level make up a layer.
![Page 4: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/4.jpg)
ANN LearningANN Learning
An ANN learns when the weights of the links are adjusted.
Learning is complete when desired output is
very close to actual output.
Learning takes place when desired output
and actual output are compared.– The difference between the two is measured
and adjustments are made to the weights
inside the ANN.
![Page 5: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/5.jpg)
Network StructuresNetwork Structures
Feedforward– Perceptrons– Multilayered
Feedback– Recurrent
![Page 6: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/6.jpg)
Feedforward SystemsFeedforward Systems
Links are unidirectional
Acyclic
In a typical, layered Feedforward network,
each unit is linked only to units in the next
layer.
![Page 7: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/7.jpg)
PerceptronsPerceptrons
Single layered networks.
Perceptron learning is very easy.
However, only linearly separable functions
can be represented by perceptrons.
![Page 8: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/8.jpg)
Optimal Linear Associative MemoryOptimal Linear Associative Memory
Architecture: Single layer Feedforward System.
![Page 9: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/9.jpg)
MultilayerMultilayer
Contain one or more layers of “hidden” nodes.
Not limited to Linearly Separable functions.
Can learn any function
There exists a popular method for learning:
back-propagation.
![Page 10: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/10.jpg)
Maxnet-Hamming NetworkMaxnet-Hamming Network
Architecture: Feedforward Multilayer System
![Page 11: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/11.jpg)
Feedback SystemsFeedback Systems
Cyclic
Output can be directed back as inputs to
previous or same level nodes.
Much more complex than a Feedforward
system.
A Recurrent system is simply a Feedback
system with closed loops.
![Page 12: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/12.jpg)
Adaptive Resonance TheoryAdaptive Resonance Theory
Architecture: Bi-directional Feedback System
![Page 13: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/13.jpg)
ApplicationsApplications
Handwriting Character Recognition
English Text Pronunciation
Driving– ALVINN (Automated Land Vehicle In a Neural Network):
requires 5 minutes of watching a human drive, and 10
minutes of back-propagation. Can drive at speeds of up to
70 mph for about 90 minutes.
Classification
![Page 14: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/14.jpg)
Multilayer vs. PerceptronMultilayer vs. Perceptron
Perceptron learns the fastest.
Perceptrons have a limited learning capacity
Perceptron is too simple for most practical
applications.
![Page 15: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/15.jpg)
Multilayer vs. FeedbackMultilayer vs. Feedback
A Feedback System more closely models the brain.
Both systems can learn any formula.
A Multilayer system uses less overhead.
Feedback Systems are a lot more complex.
![Page 16: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/16.jpg)
Most Efficient ANNMost Efficient ANN
Multilayered Feedforward system
– Relative simplicity
– Learning Capacity
– The Back-Propagation Algorithm is very good.
![Page 17: Artificial Neural Networks An Overview and Analysis](https://reader035.vdocuments.site/reader035/viewer/2022071718/56649e8e5503460f94b91f3a/html5/thumbnails/17.jpg)
The EndThe End
Any Questions?