University of Patras
Μη γραμμικές μέθοδοι συζυγών κλίσεων για βελτιστοποίηση και εκπαίδευση νευρωνικών δικτύων
Abstract
dc:descriptionThe contribution of this thesis focuses on the development and the Mathematical foundation of new conjugate gradient methods for optimization and on the study of new neural network training algorithms and their applications. We propose two new conjugate gradient methods which are based on modified new secant equations with strong theoretical advantages. Moreover, they have the attractive property of ensuring sufficient descent, avoiding thereby the usual inefficient restarts. Further, we established the global convergence of our proposed methods for general functions. The second part of the thesis is dedicated on the development of new neural network training algorithms. We propose some new conjugate gradient training methods which guarantee sufficient descent and are globally convergent for general functions. Encouraging numerical experiments on famous benchmarks verify that our proposed methods provide fast, stable and reliable convergence, outperforming classical training methods.
Degree
thesis:*- Grantor dc:publisher
- University of Patras
- Year dc:date
- 2012
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Livieris, Ioannis
- Λιβιέρης, Ιωάννης
Subjects
dc:subject × 14- Μέθοδοι συζυγών κλίσεων
- Βελτιστοποίηση χωρίς περιορισμούς
- Νευρωνικά δίκτυα
- Αλγόριθμοι εκπαίδευσης
- Conjugate gradient methods
- Unconstrained optimization
- Neural networks
- Training algorithms
- Φυσικές Επιστήμες
- Μαθηματικά
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Natural Sciences
- Mathematics
- Computer and Information Sciences
Rights
- Language dc:language
- gre
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/27470