Reconfigurable fault tolerant neural network

A reconfigurable fault-tolerant multilayer feedforward neural network is presented. It addresses both fault detection and reconfiguration of a neural network in a unified manner. As a result, a neural network with low area overhead and high reliability is achieved. A concurrent error detection scheme on neural networks which consumes low hardware overhead and negligible time overhead is described. It is capable of detecting a single fault in each neuron of the network. The reconfiguration of this neural network is addressed. It features low complexity of interconnection and low hardware overhead.<>