Part 3: Autonomous Agents 11/5/04 Positive Coupling Negative Coupling • Positive sense (sign) • Large strength 11/5/04 • Negative sense (sign) • Large strength 1 11/5/04 2 State = –1 & Local Field < 0 Weak Coupling • Either sense (sign) • Little strength h<0 11/5/04 3 11/5/04 4 1 Part 3: Autonomous Agents 11/5/04 State = –1 & Local Field > 0 State Reverses h>0 h>0 11/5/04 11/5/04 5 11/5/04 6 State = +1 & Local Field > 0 State = +1 & Local Field < 0 h>0 h<0 7 11/5/04 8 2 Part 3: Autonomous Agents 11/5/04 Hopfield Net as Soft Constraint Satisfaction System State Reverses • States of neurons as yes/no decisions • Weights represent soft constraints between decisions h<0 – hard constraints must be respected – soft constraints have degrees of importance • Decisions change to better respect constraints • Is there an optimal set of decisions that best respects all constraints? 11/5/04 9 10 Quantifying “Tension” Convergence • Does such a system converge to a stable state? • Under what conditions does it converge? • There is a sense in which each step relaxes the “tension” in the system • But could a relaxation of one neuron lead to greater tension in other places? 11/5/04 11/5/04 11 • If wij > 0, then si and sj want to have the same sign (si sj = +1) • If wij < 0, then si and sj want to have opposite signs (si sj = –1) • If wij = 0, their signs are independent • Strength of interaction varies with |wij| • Define disharmony (“tension”) Dij between neurons i and j: Dij = – si w ij s j Dij < 0 they are happy Dij > 0 they are unhappy 11/5/04 12 3 Part 3: Autonomous Agents 11/5/04 Review of Some Vector Notation Total Energy of System x1 T x = M = ( x1 L x n ) xn The “energy” of the system is the total “tension” (disharmony) in it: E {s} = Dij = si w ij s j ij n x T y = xi yi = x y = 12 si w ij s j x1 y1 xy = M x m y1 = 12 si w ij s j j x T My = = 12 sT Ws 11/5/04 13 Another View of Energy = = n j=1 x i M ij y j 14 E = E {s} E {s} = siw ij sj + si w ij s j ij ij = sk w kj s j + sk w kj s j s w s i (quadratic form) • Suppose that neuron k changes state • Change of energy: ij j jk j jk = ( sk sk ) w kj s j s h i i i = sk hk 1 T 2 = s h 11/5/04 i=1 j i 1 2 m (outer product) 11/5/04 E {s} = 12 si w ij s j 1 2 L Do State Changes Decrease Energy? The energy measures the number of neurons whose states are in disharmony with their local fields (i.e. of opposite sign): i x1 y n M xm yn L O T ji i (inner product) i=1 ij i (column vectors) 15 11/5/04 <0 jk 16 4 Part 3: Autonomous Agents 11/5/04 Proof of Convergence in Finite Time Energy Does Not Increase • In each step in which a neuron is considered for update: E{s(t + 1)} – E{s(t)} 0 • Energy cannot increase • Energy decreases if any neuron changes • Must it stop? 11/5/04 17 • There is a minimum possible energy: – The number of possible states s {–1, +1}n is finite – Hence Emin = min {E(s) | s {±1}n} exists • Must show it is reached in a finite number of steps 11/5/04 Steps are of a Certain Minimum Size If hk > 0, then (let hmin = min of possible positive h) n hk minh h = w kj s j s {±1} h > 0 = df hmin jk E = sk hk = 2hk 2hmin If hk < 0, then (let hmax = max of possible negative h) n hk maxh h = w kj s j s {±1} h < 0 = df hmax jk 18 Conclusion • If we do asynchronous updating, the Hopfield net must reach a stable, minimum energy state in a finite number of updates • This does not imply that it is a global minimum E = sk hk = 2hk 2hmax 11/5/04 19 11/5/04 20 5 Part 3: Autonomous Agents 11/5/04 Lyapunov Functions Example Limit Cycle with Synchronous Updating • A way of showing the convergence of discreteor continuous-time dynamical systems • For discrete-time system: – need a Lyapunov function E (“energy” of the state) – E is bounded below (E{s} > Emin) – E < (E)max 0 (energy decreases a certain minimum amount each step) – then the system will converge in finite time w>0 w>0 • Problem: finding a suitable Lyapunov function 11/5/04 21 The Hopfield Energy Function is Even • A function f is odd if f (–x) = – f (x), for all x • A function f is even if f (–x) = f (x), for all x • Observe: 11/5/04 22 Conceptual Picture of Descent on Energy Surface E {s} = 12 (s) T W(s) = 12 sT Ws = E {s} 11/5/04 23 11/5/04 (fig. from Solé & Goodwin) 24 6 Part 3: Autonomous Agents 11/5/04 Energy Surface + Flow Lines Energy Surface 11/5/04 (fig. from Haykin Neur. Netw.) 25 11/5/04 Flow Lines (fig. from Haykin Neur. Netw.) 27 26 Bipolar State Space Basins of Attraction 11/5/04 (fig. from Haykin Neur. Netw.) 11/5/04 28 7 Part 3: Autonomous Agents 11/5/04 Basins in Bipolar State Space energy decreasing paths 11/5/04 29 8