PERTURBATION METHOD FOR DELETING REDUNDANT INPUTS OF PERCEPTRON NETWORKS

Multilayer feedforward networks are often used for modeling complex fu nctional relationships between data sets. Should a measurable redundan cy in training data exist, deleting unimportant data components in the training sets could lead to smallest networks due to reduced-size dat a vectors. This reduction can be achieved by analyzing the total distu rbance of network outputs due to perturbed inputs. The search for redu ndant input data components proposed in the paper is based on the conc ept of sensitivity in linearized models. The mappings considered are R (I) --> R(K) with continuous and differentiable outputs, Criteria and algorithm for inputs' pruning are formulated and illustrated with exam ples.