A unified design method which generates a fault tolerant MLP while minimizing the quantization effects

We earlier presented a method to design a fault tolerant MLP which satisfies a desired level of robustness. We also separately presented an effective method to design an MLP which minimizes the effects of the quantization of the weights and the activation function caused by its hardware implementation with digital technology after the off-line learning. They are two of the critical problems which should be dealt with when we try to implement an MLP with hardware. In this paper we present a unified method which designs a fault tolerant MLP while minimizing the quantization effects.