Analysis of hidden nodes for multi-layer perceptron neural networks

摘要

The fault tolerance of the multi-layer perceptron is strongly related to its redundant hidden nodes. Increasing redundant hidden nodes may provide the advantage of higher degree of fault tolerance but have the disadvantage of higher computation and hardware complexities. The analysis of the relationship between the hidden nodes and the fault tolerance of the multi-layer perceptron is given. Fault tolerance is first discussed. Then, a new strategy for the reduction of redundant hidden nodes is proposed. It can give the maximum number of crashed nodes allowed within a given error for the neural network. In addition, some redundant hidden nodes can be removed for simple computation. Simulation results are also given to verify the method.