Fault tolerance of feedforward artificial neural networks- a framework of study

Abstract

Feedforward artificial neural networks (FFANNs) are a realization of the supervised learning paradigm. With the availability of hardware implementation of these networks, it has become desirable to measure their fault-tolerance to structural and environmental faults as well as tolerance to noise in the system variables. In this paper, the learning system model is used to describe a framework in which these studies can be conducted. Fault models are describes and error measures suggested. The relation between fault-tolerance and the generalization capabilities of the network is conjectured and the relevance of regularization capabilities of the network is conjectured and the relevance of regularization scheme to fault tolerance property discussed. The available literature on fault-tolerance of neural networks is briefly summarized in the proposed framework. Areas for further exploration are identified.