An Approximate Backpropagation Learning Rule for Memristor Based Neural Networks Using Synaptic Plasticity

摘要

We describe an approximation to backpropagation algorithm for training deepneural networks, which is designed to work with synapses implemented withmemristors. The key idea is to represent the values of both the input signaland the backpropagated delta value with a series of pulses that triggermultiple positive or negative updates of the synaptic weight, and to use themin operation instead of the product of the two signals. In computationalsimulations, we show that the proposed approximation to backpropagation is wellconverged and may be suitable for memristor implementations of multilayerneural networks.