Curvature-driven smoothing in back propagation neural networks
The standard backpropagation learning algorithm for feedforward networks aims to minimise the mean square error defined over a set of training data. This form of error measure can lead to the problem of over-fitting in which the network stores individual data points from the training set, but fails to generalise satisfactorily for new data points. In this paper is proposed a modified error measure which can reduce the tendency to overfit and whose properties can be controlled by a single scalar parameter. The new error measure depends both on the function generated by the network and on its derivatives. A new learning algorithm is derived which can be used to minimise such error measures.
Bibliographic Reference: Report: CLM-P880 EN (1990) 11pp.
Availability: Available from the Librarian, UKAEA, Culham Laboratory, Abingdon, Oxon. OX14 3DB (GB)
Record Number: 199010995 / Last updated on: 1994-12-01
Original language: en
Available languages: en