170 likes | 294 Views
Lecture 2 ASSOCIATIONS, RULES, AND MACHINES Victor Eliashberg Consulting professor, Stanford University, Department of Electrical Engineering. Slide 1. SCIENTIFIC / EGINEERING APPROACH. External system ( W,D ). Computing system, B , simulating the work of human nervous system.
E N D
Lecture 2 ASSOCIATIONS, RULES, AND MACHINES Victor Eliashberg Consulting professor, Stanford University, Department of Electrical Engineering Slide 1
SCIENTIFIC / EGINEERING APPROACH External system (W,D) Computing system, B, simulating the work of human nervous system Sensorimotor devices, D B D W Human-like robot (D,B) External world, W “When you have eliminated the impossible, whatever remains, however improbable, must be the truth.” (Sherlock Holmes) Slide 2
ZERO-APPROXIMATION MODEL s(ν) s(ν+1) Slide 3
BIOLOGICAL INTERPRETATION Working memory, episodic memory, and mental imagery Motor control AM AS Slide 4
y X X11 X12 AM sel y NM.y 0 1 NM symbol read move type symbol Teacher current state of mind next state of mind PROBLEM 1:LEARNING TO SIMULATE the Teacher This problem issimple: system AM needs to learn a manageable number of fixed rules. Slide 5
PROBLEM 2: LEARNING TO SIMULATE EXTERNAL SYSTEMThis problem ishard: the number of fixed rules needed to represent a RAM with n locations explodes exponentially with n. y 1 2 NS NOTE. System (W,D) shown in slide 3 has the properties of a random access memory (RAM). Slide 6
Programmable logic array (PLA): a logic implementation of a local associative memory (solves problem 1 from slide 5) Slide 7
BASIC CONCEPTS FROM THE AREA OF ARTIFICIAL NEURAL NETWORKS Slide 8
Typical neuron Neuron is a very specialized cell. There are several types of neurons with different shapes and different types of membrane proteins. Biological neuron is a complex functional unit. However, it is helpful to start with a simple artificial neuron (next slide). Slide 9
Neuron as the first-order linear threshold element: Inputs:xk R’ Parameters: g1,… gm R’ R’is the set of real non-negativenumbers Output: yR’ xk xm Equations: x1 gk m du g1 Σgkxk gm τ + u = (1) dt k=1 u y=L( u ) (2) where, { u if u > 0 (3) y L( u) = 0 otherwise A more convenient notation xk is the k-th component of input vector g1 x1 gk is the gain (weight) of the k-th synapse gk xk m gm y=L( u ) xm Σgkxk is the total postsynaptic current s = s k=1 τ u is the postsynaptic potential u y is the neuron output u 0 τis the time constant of the neuron y Slide 10
Input synaptic matrix, input long-term memory (ILTM) and DECODING ILTM gx1k gxnk gxik x1 xk DECODING (computing similarity) x xm s1 si sn si sn s1 An abstract representation of (1): m Σgxikxk (2) fdec: X × Gx S si = (1) i=1,…n k=1 Notation: x=(x1, .. xm)are thesignals from input neurons (not shown) gx = (gxik) i=1,…n, k=1,…m is the matrix of synaptic gains -- we postulate that this matrix represents input long-term memory (ILTM) s=(s1, .. sn)is thesimilarity function Slide 11
Layer with inhibitory connections as the mechanism of the winner-take-all (WTA) choice s1 si sn xinh q α α α ui un u1 Equations: τ τ τ (1) β β β dn di d1 (2) Note. Small white and black circles represent excitatory andinhibitorysynapses, respectively. (3) s1 sn si Procedural representation: RANDOM CHOICE iwin : { i / si=max sj > 0 } (4) ( j ) if (i == iwin) di=1; else di=0; (5) iwin “: “ denotes random equally probable choice Slide 12
Output synaptic matrix, output long-term memory (OLTM) and ENCODING di dn d1 dn d1 di y1 y ENCODING (data retrieval) yk gyki gykn gyk1 yp OLTM An abstract representation of (1): n Σgykidi yk = (2) fenc: D × Gy Y (1) k=1,…p i=1 NOTATION: d=(d1, .. dm)signals from the WTA layer (see previous slide) gy = (gyki) i=1,…n, k=1,…m is the matrix of synaptic gains -- we postulate that this matrix represents output long-term memory (OLTM) y=(y1, .. yp) output vector Slide 13
A neural implementation of a local associative memory (solves problem 1 from slide 5)(WTA.EXE) addressing by content DECODING S21(I,j) S21(i,j) Input long-term memory (ILTM) N1(j) RANDOM CHOICE Output long-term memory (OLTM) ENCODING retrieval Slide 14
A functional model of the previous network[7],[8],[11] (WTA.EXE) (1) (2) (3) (4) (5) Slide 15
Representation of local associative memory in terms of three “one-step” procedures: DECODING, CHOICE, ENCODING Slide 17