Gerl, F. and Krey, Uwe (1995) A Kuhn-Tucker cavity method for generalization with applications to perceptrons with Ising and Potts neurons. JOURNAL OF PHYSICS A-MATHEMATICAL AND GENERAL, 28 (23). pp. 6501-6516. ISSN 0305-4470
Full text not available from this repository.Abstract
Within the framework of statistical physics, we derive a cavity method for generalization by perceptrons, where the Kuhn-Tucker conditions for optimal stability are built into the cavity fields. In this way, the calculation of the generalization ability for learning processes leading to optimal stability is simplified. Within our approach, the degrees of freedom of the neurons can be rather arbitrary. For perceptrons with Ising neurons we relate our method to the traditional replica approach. New results are obtained for Q-state Ports model perceptrons, including the asymptotic behaviour for alpha --> infinity and general Q.
| Item Type: | Article |
|---|---|
| Uncontrolled Keywords: | ALGORITHM; ADATRON |
| Subjects: | 500 Science > 530 Physics |
| Divisions: | Physics > Institute of Theroretical Physics |
| Depositing User: | Dr. Gernot Deinzer |
| Date Deposited: | 15 Nov 2023 10:36 |
| Last Modified: | 15 Nov 2023 10:36 |
| URI: | https://pred.uni-regensburg.de/id/eprint/52165 |
Actions (login required)
 |
View Item |
