AN ADAPTIVE FAULT TOLERANT NEURAL NETWORK CLASSIFIER

Based on the research of fault tolerant neural networks by some precursors, an adaptive fault tolerant neural networks learning algorithm FTART4 is proposed to resolve the multi node open fault problem. FTART4 uses a distinct mechanism to deal with the dilemma of fault tolerance and redundance by ascertaining a fault sensitive parameter. The problem’s formalization and the specific FTART4 algorithm are given. By applying the FTART4 to some standard benchmarks, the experitments show that this method achieves preferable results.