MEMRISTOR-BASED NEURAL NETWORK TRAINING METHOD AND TRAINING APPARATUS THEREOF

摘要

A memristor-based neural network training method and a training apparatus thereof. A neural network comprises a plurality of neuron layers which are connected one by one and weight parameters among the neuron layers. The training method comprises: training the weight parameters of the neural network, and programming a memristor array on the basis of the trained weight parameters so as to write the trained weight parameters into the memristor array; and updating at least one layer of weight parameters of the neural network by adjusting part of conductance values of the memristor array. According to the training method, the defects of implementation schemes of on-chip training and off-chip training of the memristor neural network are overcome; from the perspective of realization of the neural network system, function degradation of the neural network system caused by non-ideal characteristics of devices such as a yield problem, a non-consistency problem, conductance drift and random volatility is solved, the complexity of the neural network system is greatly simplified, and the realization costs of the neural network system are reduced.