In this paper we review recent theoretical approaches for analysing the dynamics of on-line learning in multilayer neural networks using methods adopted from statistical physics. The analysis is based on monitoring a set of macroscopic variables from which the generalisation error can be calculated. A closed set of dynamical equations for the macroscopic variables is derived analytically and solved numerically. The theoretical framework is then employed for defining optimal learning parameters and for analysing the incorporation of second order information into the learning process using natural gradient descent and matrix-momentum based methods. We will also briefly explain an extension of the original framework for analysing the case where training examples are sampled with repetition.
|Title of host publication||Exploratory Data Analysis in Empirical Research: Proceedings of the 25th Annual Conference of the Gesellschaft für Klassifikation e.V., University of Munich, March 14-16, 2001|
|Editors||Manfred Schwaiger, Otto Opit|
|Number of pages||10|
|Publication status||Published - 2003|
|Event||Studies in Classification, Data Analysis and Knowledge Organization - |
Duration: 1 Jan 2003 → 1 Jan 2003
|Name||Studies in Classification, Data Analysis, and Knowledge Organization|
|Other||Studies in Classification, Data Analysis and Knowledge Organization|
|Period||1/01/03 → 1/01/03|
Bibliographical noteThe original publication is available at www.springerlink.com
- on-line learning
- neural networks
- statistical physics
- natural gradient descent
Saad, D. (2003). The theory of on-line learning: a statistical physics approach. In M. Schwaiger, & O. Opit (Eds.), Exploratory Data Analysis in Empirical Research: Proceedings of the 25th Annual Conference of the Gesellschaft für Klassifikation e.V., University of Munich, March 14-16, 2001 (pp. 300-309). (Studies in Classification, Data Analysis, and Knowledge Organization). Springer.