Dual channel and dual feedback loop self-learning memristive neural network circuit and its application.

Abstract

Elman neural network has a local memory function and is widely used in engineering applications. However, there was no previous literature on the hardware circuit implementation of Elman neural network. In this paper, a dual channel and dual feedback loop self-learning memristive circuit based on Elman neural network and Hebbian learning rule is discussed. Firstly, a single neural network model has been proposed. Then, a single neural network self-learning memristive circuit with a local feedback loop and a global feedback loop is designed. The simulation results show that the introduction of the local feedback loop can accelerate the self-learning process. Secondly, to suppress the overlearning situation, an upgraded single neural network self-learning memristive circuit is proposed. Here, the Hebbian learning rule is embedded into the global feedback loop based on the original circuit. Thirdly, a multiple neural network self-learning memristive circuit is designed based on the upgraded circuit, which is applied to the recognition of 5 × 3 pixel number images and grayscale images. It shows a fast speed and a high accuracy during the recognition process, which verifies the feasibility of this work.