Application of Two-Stage Learning on Brain-like System

Abstract

This paper proposes a two-stage learning strategy constructed by two kinds of neural networks to simulate function of human brain. In the first stage, sequence images from environment input to a HOSM neural network. By unsupervised learning, the weights are fixed which can extract local features like vision's receptive field. In the second stage, an improved HDR neural network is built by supervised learning. This proposed structure has been implemented on a brain-like robot. Experimental results show that the learning strategy is effect.