Limits to the fault-tolerance of a feedforward neural network with learning

Input data and hardware fault tolerance of neural networks are discussed. It is shown that fault-tolerant behavior is not self-evident but must be activated by an appropriate learning scheme. Practical limitations are demonstrated by an example of neural character recognition. The results show that the effects of learning and synapse weight decay on fault tolerance largely influence the practicality of large-scale silicon implementations. It is anticipated that, owing to implementation issues, such as the use of volatile memories, some neural VLSI architectures will not be sufficiently fault tolerant.