Abstract
We present a method for determining the globally optimal on-line learning rule for a soft committee machine under a statistical mechanics framework. This work complements previous results on locally optimal rules, where only the rate of change in generalization error was considered. We maximize the total reduction in generalization error over the whole learning process and show how the resulting rule can significantly outperform the locally optimal rule.
| Original language | English |
|---|---|
| Pages (from-to) | 322-328 |
| Number of pages | 7 |
| Journal | Advances in Neural Information Processing Systems |
| Volume | 10 |
| Publication status | Published - Jan 1998 |
| Event | Advances in Neural Information Processing Systems 1994 - Singapore, Singapore Duration: 16 Nov 1994 → 18 Nov 1994 |
Bibliographical note
Copyright of the Massachusetts Institute of Technology Press (MIT Press)Keywords
- on-line learning
- statistical mechanics
- generalization error
- optimal rule
- resulting rule
Fingerprint Dive into the research topics of 'Globally optimal on-line learning rules'. Together they form a unique fingerprint.
Cite this
Rattray, M., Saad, D., Jordan, M. I. (Ed.), Kearns, M. J. (Ed.), & Solla, S. A. (Ed.) (1998). Globally optimal on-line learning rules. Advances in Neural Information Processing Systems, 10, 322-328.