computational modeling of neural networks and memory simulation
DESCRIPTION
Computational Modeling of Neural Networks and Memory Simulation. Alex Sonal Carl Ashwin Rebecca Shreyas Madhu Seth Jeff James Jonathan . Overview: Background Inspirations Biology and Neuroscience Computer Modeling Project Design and Coding Successes and Challenges - PowerPoint PPT PresentationTRANSCRIPT
Overview:•Background• Inspirations• Biology and Neuroscience• Computer Modeling
•Project• Design and Coding• Successes and Challenges
•Central Question: Can we model a human brain?
Computational Modeling of Neural Networks and Memory Simulation
NJ Governor’s School for the SciencesTeam Project T7Dr. Minjoon KouhAaron Loether
AlexSonalCarlAshwinRebeccaShreyas
MadhuSethJeffJamesJonathan
Current State of Neuroscience
◦ Anatomy is well understood◦ Lack of a cohesive brain theory Emergent properties Prediction versus Behavior
The Brain
MemoryBiology
◦ Patterns of active and inactive neurons in neural networks
◦ Vividness is determined by interneuron connection strength
Psychology◦ Forgetting◦ “Networks of
knowledge” (associative memory)
The Hebbian Theory“Neurons that Fire Together, Wire Together”If activity of two neurons is correlated strong synaptic
connection
ON
ON
OFF
Strong Weak
The Hopfield NetworkHopfield Network
If stimulus activates single neuron, other related neurons in neural network will also become activated
NJGSSScience
s
School
Research
Friends
Projects
(Input)(Output
)
THE PROGRAMPaul
Paul Jr.Paul Jr. Jr.
Step 1: Process Images
101001...
Step 2: Memorize
W11 W12W13W14
W21 W22W23W24
W31 W32 W33W34
W41 W42W43W44
1 2
340 1
1-1
1 01-1
1 1 0-1
-1 -1-10
4 3
1 2
0 1 -1 -1
1 0 -1 -1
-1 -1 0 1
-1 -1 10
0 2 0 -2
2 0 0 -2
0 0 0 0
-2 -2 0 0
N1
N2
N3N4
N1
N2
N3N4
N1 N2 N3 N4 N1 N2 N3 N4
Step 3: Scramble
Step 4: RecallY(t) = W*Y(t-1)
Pictures Memorized vs. Accuracy of Recall
More pictures in the Memory
Performance =
Range from -1 to 1
Worse Recall
The Effect of Noise on Recall
More Noise
Worse Recall
Residual=
performance of output –
performance of input
ChallengesMemory of MATLABPicture similarity
Result: low resolution pictures and low performance
The Future of the Hopfield Model
- Brain Theory
- Artificial Intelligence
- Education
SPECIAL THANKS TO:Dr. KouhAaron LoetherHopfield and HebbDr. MiyamotoMs. Papier
BUT MOST OF ALL:Donors Who Helped Make NJGSS ‘11 Possible!!
Sources Cited• Anastasio T J. Tutorial on Neural Systems Modeling. Sunderland
(MA): Sinauer Associates Inc.; 2010. 583 p.• Gazzaniga M S. The Cognitive Neurosciences. Cambridge (MA):
Bradford; 1997. 1447 p.• Wells R B.Synaptic Weight Modulation and Adaptation. In:
University of Idaho MRCI [discussion list on the Internet]. 2003 May 15; [cited 2011 July]. 13 p. Available from: http://www.mrc.uidaho.edu/~rwells/techdocs/Synaptic%20Weight%20Modulation%20and%20Adaptation%20I.pdf
• Kandel E R. Principles of Neuroscience. New York (NY): McGraw-Hill; 2000. 1414 p.
• Dayhoff J. School of Computing [homepage on the Internet]. Leeds (UK): University of Leeds; 2003. [cited 2011]. Available from: http://www.comp.leeds.ac.uk/ai23/reading/Hopfield.pdf.