Structure of Multi-layer Neural Networks for fault tolerance

We propose a new learning algorithm to enhance fault tolerance of multi-layer neural networks (MLN). This method is based on the idea that strong weights make MLN sensitive to faults. The purpose of new learning algorithm is to remove obstructions of fault tolerance from MLNs. We discuss about obstructions (strong connection and bias of each unit). As a result, we proposed new learning algorithm which is restricting the absolute value of weight and constructing MLNs without bias. We apply this algorithm to pattern recognition problems. It is shown that the fault tolerance of MLNs is improved