Second order properties of error surfaces: Learning time and generalization

Yann LeCun, I. Kanter, Sara A. Solla

Research output: Contribution to journalArticlepeer-review

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.
Original languageAmerican English
Pages (from-to)918-924
JournalAdvances in Neural Information Processing Systems
Volume3
StatePublished - 1991

Fingerprint

Dive into the research topics of 'Second order properties of error surfaces: Learning time and generalization'. Together they form a unique fingerprint.

Cite this