Second Order Properties of Error Surfaces: Learning Time and Generalization

Part of Advances in Neural Information Processing Systems 3 (NIPS 1990)

Bibtex Metadata Paper

Authors

Yann LeCun, Ido Kanter, Sara Solla

Abstract

The learning time of a simple neural network model is obtained through an analytic computation of the eigenvalue spectrum for the Hessian matrix, which describes the second order properties of the cost function in the space of coupling coefficients. The form of the eigenvalue distribution suggests new techniques for accelerating the learning process, and provides a theoretical justification for the choice of centered versus biased state variables.