Mixed-signal VLSI implementation of the Products of Experts' contrastive divergence learning scheme

We present circuits and methods for the on-chip VLSI implementation of the "contrastive divergence" learning scheme used in the Product of Experts (PoE) architecture. A novel circuit changes a neuron's synaptic weight accurately and experimental results from its aVLSI implementation are given. Additional circuitry required to control the weight change circuit and to implement the learning rule on-chip is also presented.