Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids:AVQ Al gor i thms Unsupervised Competitive Learning (UCL) m(t+1)=(t)+cx(t)-m(t)] 6-14 i(t+1)=mi(t)fi≠j 6-15 fcdefines a slowly decreasing sequence of learning woece-o coefficient for 10,000 samples x(t) Supervised Competitive Learning (SCL) m(t+）=m()）+c(x()[x()-m()] 6-16 m0+cx0-m0】x∈D6-17 mi(t)-cix(t)-mi(t)]ifx Dj 2006.11.10 11
2006.11.10 11 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms Unsupervised Competitive Learning (UCL) ( 1) ( ) 6 15 ( 1) ( ) [ ( ) ( )] 6 14 + = − + = + − − m t m t if i j m t m t c x t m t i i j j i j {ci} defines a slowly decreasing sequence of learning coefficient for 10,000 samples ( ) 10,000 For instance , 0.1 1 x t t ci = − Supervised Competitive Learning (SCL) 6 17 ( ) [ ( ) ( )] ( ) [ ( ) ( )] ( 1) ( ) ( ( )) ( ) ( ) 6 16 − − − + − = + = + − − m t c x t m t if x Dj m t c x t m t if x Dj m t m t c r x t x t m t j i j j i j j j i j j
Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids:AVQ Al gor i thms Differential Competitive Learning (DCL) m(t+1)=m(t)+c△S((t)[x(t)-(t)] 6-18 mi(t+1)=(t)fi≠j 6-19 AS(,(t+1))denotes the time change of the jth neuron's competitive signal.In practice we only use the sign of(6-20) AS(y,(t+1)=S0t+1)-S(t) 6-20 Stochastic Equilibrium and Convergence Competitive synaptic vector converge to decision-class centroids. May converge to locally maxima 2006.11.10 12
2006.11.10 12 Chapter 6 Architecture and Equilibria 6.3 Synaptic convergence to centroids: AVQ Algorithms Differential Competitive Learning (DCL) ( 1) ( ) 6 19 ( 1) ( ) ( ( ))[ ( ) ( )] 6 18 + = − + = + − − m t m t if i j m t m t c S y t x t m t i i j j t j j j S (y (t +1)) j j denotes the time change of the jth neuron’s competitive signal . In practice we only use the sign of (6-20) Sj(yj (t +1)) = Sj(yj(t +1)) − Sj(yj(t)) 6− 20 Stochastic Equilibrium and Convergence Competitive synaptic vector converge to decision-class centroids. May converge to locally maxima