ist 4 information and logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf ·...

92
IST 4 Information and Logic

Upload: phamtram

Post on 06-Mar-2018

216 views

Category:

Documents


2 download

TRANSCRIPT

Page 1: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

IST 4Information and Logic

Page 2: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

mon tue wed thr fri30 M1 1= todayT

6 M1

13 1 2M2x= hw#x out

oh oh

oh oh

= todayT

M

20 227 M2

x= hw#x due

idt

oh oh

h h Students’ MQ27 M2

4 311 3 4Mx= MQx out

midtermsoh

oh oh

oh = office hours oh Students MQpresentations

11 3 418 4 525

Mx= MQx out

Mx= MQx dueoh oh

ohoh

25

1 5oh

oh

oh

oh

T

Page 3: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

MQsMQs1. Everyone has a gift! (Tuesday)

2. Memory (Thursday)

Page 4: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Tuesday 6/2 2:30pm –

1. Christopher Haack: The gift of resilience

Tuesday, 6/2, 2:30pm

p g

2. Joon Lee: Settling is not an option

3 Spencer Strumwasser: The gift of dyslexia 3. Spencer Strumwasser: The gift of dyslexia

4. Richard Zhu: The gift of memory

5 A h i H i Th ift f i l iti5. Ashwin Hari: The gift of musical composition

6. Jessica Nassimi: Evolution—a gift in disguise

7. Serena Delgadillo: The gift of self-expression

8. Megan Keehan: Gift of motherliness

9. Zane Murphy: Grandmother and the piano

Page 5: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Thursday 6/4 2:30pm –Thursday, 6/4, 2:30pm

1. Connor Lee: Memory is a fickle thing – blessing or curse

2. Pallavi Aggarwal: The wonders of human memory

3. Peter Kundzicz and Anshul Ramachandran: Muscle memories

4. Siva Gangavarapu: A cultural retrospection. g p p

5. Philip Liu: The light of other days

6 Jason Simon: Math and Broadway6. Jason Simon: Math and Broadway

7. Yujie Xu: Memory v.s. ESL

8 C l Zh Wh 8. Celia Zhang: When memory sours

Page 6: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Last LectureGates and circuits AON: AND, OR, Not

LT: Linear ThresholdLT: Linear Threshold

>>

abc

ab >>

>>

abc

ab >

>

>

>

bc

abc

abc

>>>>

>>

>>

bc

abc

abc

Page 7: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Last LectureAON: AND, OR, Not LT: Linear Threshold

General construction for symmetric functions

AON 54

AONLT-l

*

*

2LT-nl *

Exponential gap in sizeWhat are the symmetric functions that can be computed functions that can be computed by a single LT gate? * = it is optimal

Page 8: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Linear Threshold and SYM

Page 9: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

LT: Linear Threshold

Page 10: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Symmetric Functions and LT Circuits

Q: How is SYM related to LT1 ??

Q: Which class has more functions?

Definitions:

Q

Definitions:

(1) SYM = the class of Boolean symmetric functions

(2) LT1 = the class of Boolean functions that can be (2) LT1 the class of Boolean functions that can be realized by a single LT gate.

Page 11: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

AND, OR, XOR and MAJ are symmetric functions

Q Whi h t i f ti i LT ?

|X| AND OR XOR MAJ

Q: Which symmetric functions are in LT1?

0 0 0 0 0

1 0 1 1 01 0 1 1 0

2 0 1 0 1

3 1 1 1 1

LT1 LT1 LT1not LT1

LT1 = the class of Boolean functions that can be realized by a single LT gate.

LT1 LT1 LT1not LT1

realized by a single LT gate.

Page 12: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Definition: A t i B l f ti i i TH if it h t t A symmetric Boolean function is in TH if it has at most a single transition in the symmetric function table

= a transition

|X| AND OR XOR MAJ

0 0 0 0 0

1 0 1 1 0

2 0 1 0 1

3 1 1 1 1

Not in THIn TH

Page 13: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

The Class TH is in LT1

|X| TH0 TH1 TH2 TH3 TH0 TH1 TH2 TH3

0 1 0 0 0 0 1 1 10 1 0 0 0 0 1 1 11 1 1 0 0 0 0 1 1

1 1 1 0 0 0 0 12 1 1 1 0 0 0 0 1

3 1 1 1 1 0 0 0 0

Page 14: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Q: How is TH related to SYM and LT1 ??

We know that:

Q

We know that:

We Proved that:

SYM LT1TH

Page 15: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

TH is exactly in the intersection of SYM and LT1

Theorem:

Q: What are the 4 functions?

Proof: Not today... you might want to try and prove it...

Q: What are the 4 functions?

SYM LT1TH

???

Page 16: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

LT1 Function that is not SymmetricLT1 Function that is not Symmetric

0 00 1

-1-2

00

11

1 01 1

0-1

10

-1-1

Page 17: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

lLinear Threshold Circuits

for symmetric functionsfor symmetric functions

Page 18: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

5AON 54

AONLT-l

L l 2LT-nl

General construction for symmetric functions

Page 19: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

|X| XOR Q: compute XOR with TH gates?

0 01 1

with TH gates?

2 0

1|X| TH1 TH2 TH1+TH2-1

0 0 1 01 1 1 12 1 0 02 1 0 0

Page 20: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

LT Depth-2 Circuits

TH1

+-1

TH2 |X| TH1 TH2 TH1+TH2-1 |X| TH1 TH2

0 0 1 01 1 1 11 1 1 12 1 0 0

Page 21: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Generalization

|X| f( )

z

|X| f(x)0 0

1 1

2 1

3 0

4 04 0

Page 22: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Generalization

|X| f( )

z

|X| f(x)0 0

1 1

2 1

3 0

4 04 0

Page 23: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Generalization

|X| f( ) TH1

z

|X| f(x) TH1

0 0 0

1 1 1

2 1 1

3 0 1

4 0 14 0 1

Page 24: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Generalization

|X| f( ) TH1 TH3

z

|X| f(x) TH1 TH3

0 0 0 1

1 1 1 1

2 1 1 1

3 0 1 0

4 0 1 04 0 1 0

Page 25: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Generalization

|X| f( ) TH1 TH3 Σ 1

z

|X| f(x) TH1 TH3 Σ -10 0 0 1 0

1 1 1 1 1

2 1 1 1 1

3 0 1 0 0

4 0 1 0 04 0 1 0 0

Page 26: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

|X| f( ) TH1 TH3 Σ 1|X| f(x) TH1 TH3 Σ -10 0 0 1 0

1 1 1 1 1

2 1 1 1 1

3 0 1 0 0

4 0 1 0 04 0 1 0 0

Page 27: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Generalization to EQz EQ

00

00 0

1

1

1

n

1

00

Page 28: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Generalization to EQz EQ

00 1

1

1

12

0

1

0 1

Page 29: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Generalization to SYMz M

+-1

Q:h h l fWhat is the generalization to arbitrary symmetric functions?

Page 30: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Generalization to SYMz M

Q:Q:What is the generalization to arbitrary symmetric functions?

A:A:Consider the symmetric function table, it is a sum of non-overlapping 1-intervals

0

0

1

1 Sum of two TH functions

Page 31: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Back to XORX

n TH gates for XOR of n variables

0 0

1

2

1

0

3

4

1

0

5 1

Page 32: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

LT l Circuit Design Algorithm for SYMLT-l Circuit Design Algorithm for SYM

f(X)

0 1

f(X)

0

1

2

1

1

02

3

0

1Subtract 1 for every

4

5

1

0

Subtract 1 for everyisolated 1-block

67

11

Page 33: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

The Layered Construction for SYM -Some History

Saburo Muroga1925- 2009

1959

Was born in Japan Majority Decision

PhD in 1958 from Tokyo U, Japan

1960-1964: Researcher at IBM Research, NY,

1964-2002: professor at the University of Illinois, Urbana-Champaign

Page 34: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Saburo Muroga1925- 2009

HW#5 problem 2a

Page 35: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

neural circuits and logicg

some more historysome more history...

Page 36: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Being Homeless and Interdisciplinary Research

W M C ll hWarren McCulloch1899 - 1969

Walter Pitts1923 - 1969

Neurophysiologist, MD Logician, Autodidactg ,

Warren McCulloch arrived in early 1942 to the University of Chicago, invited Pitts, who was homeless, to live with his family

h ll h d P ll b d In the evenings McCulloch and Pitts collaborated. Pitts was familiar with the work of Leibniz on computing.They considered the question of whether the nervous system is a kind of universal computing device as described by Leibnizuniversal computing device as described by Leibniz

This led to their 1943 seminal neural networks paper:A Logical Calculus of Ideas Immanent in Nervous Activity

Page 37: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Warren McCulloch W l Pi

ImpactWarren McCulloch

1899 - 1969 Walter Pitts1923 - 1969

Neurophysiologist, MD Logician, Autodidactg ,

This led to their 1943 seminal neural networks paper:p pA Logical Calculus of Ideas Immanent in Nervous Activity

Neural networks d Logic Ti MNeural networks and Logic Time Memory

Threshold Logic and Learning

State Machines

Page 38: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

neural circuits and memorym m y

computing with dynamicscomputing with dynamics

Page 39: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Linear ThresholdSome AdjustmentsSome Adjustments

Linear Threshold (LT) gateLinear Threshold (LT) gate

-tthreshold

t

-t

-1

t

1

-1

Page 40: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

AND Function with {0,1}

-21

-21

0 00 1

1 0

-2-11

0001 0

1 1-10

01

Page 41: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

AND Function with {-1,1}

Th AND f ti f t i bl ith { 1 1} ???The AND function of two variables with {-1, 1}: ???

-1-1-11

-3-1--

1-111

11

1-11-+

Page 42: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

AND Function with {-1,1}

Th AND f ti f t i bl ith { 1 1} ???The AND function of two variables with {-1, 1}: ???

-1-1-11

-3-1

1-111

11

1-11

Page 43: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Linear Threshold with MemoryEl h t b l f i d i A i ltElephants are symbols of wisdom in Asian culturesand are famed for their exceptional memory

A memory nosey

Remembers the last f(X)

Page 44: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksE lExample

-1 -1 00

weights thresholds

Th t t f th t k th t th t d The state of the network: the vector that corresponds to the states (noses…) of the gates

Page 45: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2Label the gates

-1 -1 00 1 1

1 2Label the gates

-1 -1 00

Page 46: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 1 1

1 2

1-1 -1 00

Page 47: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 -1 1

1 2

1-1 -1 00

Page 48: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 -1 1

1 2-1

-1 -1 00

11 i bl -11 is a stable state

Page 49: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 1 1

1 2

-1 -1 00

Page 50: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 1 1

1 21

-1 -1 00

Page 51: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 1 -1

1 21

-1 -1 00

Page 52: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 1 -1

1 2-1

-1 -1 00

1 1 i bl 1-1 is a stable state

Page 53: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 1 1

1 2

-1 -1 00

State transition diagram (state space)1

The node

11 -11

2

1state

The node that computes -1-11-1

2Q:Is -1-1 a stable state?

Page 54: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 -1 -1

1 2-1

-1 -1 00

Answer: NoAnswer: No

Q:Is -1-1 a stable state?

Page 55: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 1 -1

1 2-1

-1 -1 00

11 111

11 -11

2-1-11-1

1

Page 56: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 -1 -1

1 2-1

-1 -1 00

11 111

11 -11

2-1-11-1

1

Page 57: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 -1 1

1 2-1

-1 -1 00

11 111

11 -11

2 2-1-11-1

1

Page 58: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksExampleExample

1 2

-1 -1 00 -1 1

1 2

-1 -1 00

11 111stable states

11 -11

2 2-1-11-1

1

Page 59: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

neural circuits and memorym m y

associative memoryassociative memory

Page 60: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksComputing with DynamicsComputing with Dynamics

1stable states11 -11

2

1

2

stable states

-1-11-1

2

1

2

Input:initial state

Output:stable state

FeedbackNetwork

n t al state

11 -1111 11

Page 61: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksComputing with DynamicsComputing with Dynamics

1stable states11 -11

2

1

2

stable states

-1-11-1

2

1

2

Input:initial state

Output:stable state

FeedbackNetwork

n t al state

11 1-111 1 1

Page 62: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksComputing with DynamicsComputing with Dynamics

Input: Output:Associative Memory“The Leibniz-Boole Machine”p

initial statep

stable stateThe Leibniz Boole Machine

FeedbackNetwork

Page 63: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksComputing with DynamicsComputing with Dynamics

Input: Output:Associative Memory“The Leibniz-Boole Machine”p

initial statep

stable stateThe Leibniz Boole Machine

FeedbackNetwork

Page 64: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksComputing with DynamicsComputing with Dynamics

Input: Output:Associative Memory“The Leibniz-Boole Machine”p

initial statep

stable stateThe Leibniz Boole Machine

FeedbackNetwork

Page 65: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksComputing with DynamicsComputing with Dynamics

Input: Output:Associative Memory“The Leibniz-Boole Machine”p

initial statep

stable stateThe Leibniz Boole Machine

FeedbackNetwork

Page 66: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksComputing with DynamicsComputing with Dynamics

Input: Output:Associative Memory“The Leibniz-Boole Machine”p

initial statep

stable stateThe Leibniz Boole Machine

FeedbackNetwork

Page 67: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksComputing with DynamicsComputing with Dynamics

Input: Output:Associative Memory“The Leibniz-Boole Machine”p

initial statep

stable stateThe Leibniz Boole Machine

FeedbackNetwork

Page 68: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksComputing with DynamicsComputing with Dynamics

Input: Output:Associative Memory“The Leibniz-Boole Machine”p

initial statep

stable stateThe Leibniz Boole Machine

FeedbackNetwork

Page 69: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksComputing with DynamicsComputing with Dynamics

Input: Output:Associative Memory“The Leibniz-Boole Machine”p

initial statep

stable stateThe Leibniz Boole Machine

FeedbackNetwork

Page 70: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

????

Who is this person?????

Page 71: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

John Hopfield

f ld d l ( l h )Feedback Networks

Hopfield Model (Caltech 1982)

Page 72: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback NetworksJohn Hopfield

f ld d l ( l h )

1 2

Hopfield Model (Caltech 1982)

-1 -1 00 1 1

-11 20 0

1 1

1

i = node i 0 = threshold ti

-1

1 = state vi -1 = weight of edge (i,j)

Page 73: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

The matrix descriptionm p

Page 74: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Feedback Networks/The Vector/Matrix Description

An n node feedback network can be specified by:An n node feedback network can be specified by:• W an nxn matrix of weights• T an n vector of thresholds f• V an n vector of states

1

4

2

34

5

Page 75: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

The Matrix DescriptionE lExample

An n node feedback network can be specified by:An n node feedback network can be specified by:• W an nxn matrix of weights• T an n vector of thresholds f• V an n vector of states

1 20 0

-121

-1

11 1

-1

-1

Page 76: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

The Matrix DescriptionComputationComputation

C t ti i N (W T)Computation in N= (W,T)1

4

2

34

5by column

Page 77: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Order of computationf mp

serial and parallel

Page 78: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Modes of OperationModes of Operation

Q: when do the nodes compute?Q: when do the nodes compute?

Serial mode: one node at a time (arbitrary order)

-1

( y )

21-1

Page 79: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Modes of OperationModes of Operation

Q: when do the nodes compute?Q: when do the nodes compute?

Serial mode: one node at a time (arbitrary order)

-1

( y )

21-1

Page 80: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Modes of OperationModes of Operation

Q: when do the nodes compute?Q: when do the nodes compute?

Serial mode: one node at a time (arbitrary order)

-1

( y )

21-1

Fully-Parallel mode: all nodes at the same time-1

21-1

Page 81: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Three examplesmp

Page 82: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Example 1Serial Mode – Symmetric Weight MatrixSerial Mode – Symmetric Weight Matrix

21-1

-11

The state space:

1stable states

The state space:

11 -11

2

1

2-1-11-1

1

Page 83: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Example 2Fully Parallel (FP) Mode Symmetric Weight MatrixFully-Parallel (FP) Mode – Symmetric Weight Matrix

Q: how does the state space look?

21-1

Q p

21-1

start with 11start with 11

It’s acycle!cycle!

Page 84: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Example 2Fully Parallel (FP) Mode Symmetric Weight MatrixFully-Parallel (FP) Mode – Symmetric Weight Matrix

21-1

21-1

h The state space:

stable states11 -11

l f l th 2-1-11-1

cycle of length 2

Page 85: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Example 2Fully Parallel Mode Symmetric Weight Matrix

3AntisymmetricFully-Parallel Mode – Symmetric Weight MatrixAntisymmetric

W T = −W21

-1

1

11W W

-1

Q: how does the state space look?p

Page 86: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Example 3Fully Parallel Mode Antisymmetric Weight Matrix

1

Fully-Parallel Mode – Antisymmetric Weight Matrix

21-1

Q: how does the state space look?

cycle of length 4

Page 87: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Example 3Fully Parallel Mode Antisymmetric Weight Matrix

1

Fully-Parallel Mode – Antisymmetric Weight Matrix

21-1

The state space:

11 -11

-1-11-1

cycle of length 4

Page 88: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

The Three Cases

11 -1111 -11 11 -11

-1-11-1-1-11-1 -1-11-1

1 2 3

Wmode symmetric antisymmetricCycle lengths

serial 1 ?1Example #

fully-parallel 1,2 42 3

Page 89: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

The Three Cases

Wmode symmetric antisymmetricCycle lengths

mode ymm ymm

serial 1 ?1

Example #fully-parallel 1,2 4

2 3

Example #

1 Hopfield 1982

2 G l 19852 Goles 1985

3 Goles 1986

Page 90: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Proof Ideas

C l l th W symmetric antisymmetricCycle lengths Wmode symmetric antisymmetric

serial 1 ?1Example #

fully-parallel 1,2 4

1

2 3The proofs of these three results use the conceptof an energy function

For the serial mode:

Sh th tShow that:

Namely, stable states are local max of the energy E

Page 91: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity

Questions on Convergence

Posted on the class web site

Cycle lengths Wd symmetric antisymmetric

Posted on the class web site

Cycle lengths Wmode symmetric antisymmetric

serial 1 ?1Example #

fully-parallel 1,2 42 3

1 Hopfield 1982

2 Goles 1985 Q1: Are the three cases “distinct”?2 Goles 1985

3 Goles 1986

Q

Q2: Elementary proof? (wo/energy)

Page 92: IST 4 Information and Logic - brain.caltech.edubrain.caltech.edu/ist4/lectures/lect 1715 p.pdf · AON: AND, OR, Not LT: ... A Logical Calculus of Ideas Immanent in Nervous Activity