Obtaining Fault Tolerant Multilayer Perceptrons Using an Explicit Regularization

摘要: When the learning algorithm is applied to a MLP structure, different solutions for weight values can be obtained if parameters of rule or initial conditions are changed. Those present similar performance with respect learning, but they differ in other aspects, particular, fault tolerance against perturbations. In this paper, backpropagation that maximizes proposed. The presented explicitly adds new term related mean square error degradation presence deviations order minimize degradation. results demonstrate efficiency proposed here comparison algorithm.