Networks and Memory Simulation

trainerhungarianΤεχνίτη Νοημοσύνη και Ρομποτική

20 Οκτ 2013 (πριν από 3 χρόνια και 11 μήνες)

89 εμφανίσεις

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 Sciences

Team Project T7

Dr.
Minjoon

Kouh

Aaron
Loether

Alex

Sonal

Carl

Ashwin

Rebecca

Shreyas

Madhu

Seth

Jeff

James

Jonathan

Current State of
Neuroscience



Anatomy is well understood



Lack of a cohesive brain theory



Emergent properties



Prediction versus Behavior




The Brain

Memory


Biology


Patterns of active and
inactive neurons in
neural networks


Vividness is
determined by
interneuron
connection strength


Psychology


Forgetting


“Networks of
knowledge”

(
a
ssociative 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 Network


Hopfield Network


If stimulus activates single neuron, other related neurons
in neural network will also become activated

NJGSS



Sciences



School

Research

Friends

Projects

(Input)

(Output)

THE PROGRAM

Step 1: Process Images

1

0

1

0

0

1

.

.

.


Step 2: Memorize


W
11

W
12

W
13

W
14


W
21

W
22

W
23

W
24


W
31
W
32

W
33

W
34


W
41

W
42

W
43

W
44





1

2

3

4


0

1

1

-
1


1

0

1

-
1


1

1

0

-
1


-
1

-
1

-
1

0





4

3

1

2


0

1
-
1
-
1


1

0
-
1
-
1


-
1

-
1

0 1


-
1
-
1

1

0






0

2 0
-
2


2

0 0
-
2


0

0


0 0


-
2
-
2


0

0











N1

N2

N3

N4

N1

N2

N3

N4

N1

N2

N3

N4

N1

N2

N3

N4

Step 3: Scramble

Step 4: Recall

Y(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

Challenges


Memory of MATLAB


Picture similarity

Result: low resolution pictures and low performance

The Future of the Hopfield
Model

-

Brain Theory

-

Artificial Intelligence

-

Education

SPECIAL THANKS TO:

Dr.
Kouh

Aaron
Loether

Hopfield and
Hebb

Dr. Miyamoto

Ms.
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%20Weig
ht%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
.