Gradient descent learning in and out of equilibrium

Journal Article · Fri Jun 01 00:00:00 EDT 2001 · Physical Review E

DOI:https://doi.org/10.1103/PhysRevE.63.061905· OSTI ID:40203373

Caticha, Nestor; Araujo de Oliveira, Evaldo

Relations between the off thermal equilibrium dynamical process of on-line learning and the thermally equilibrated off-line learning are studied for potential gradient descent learning. The approach of Opper to study on-line Bayesian algorithms is used for potential based or maximum likelihood learning. We look at the on-line learning algorithm that best approximates the off-line algorithm in the sense of least Kullback-Leibler information loss. The closest on-line algorithm works by updating the weights along the gradient of an effective potential, which is different from the parent off-line potential. A few examples are analyzed and the origin of the potential annealing is discussed.

• Sponsoring Organization: (US)
• OSTI ID: 40203373
• Journal Information: Physical Review E, Vol. 63, Issue 6; Other Information: DOI: 10.1103/PhysRevE.63.061905; Othernumber: PLEEE8000063000006061905000001; 109105PRE; PBD: Jun 2001; ISSN 1063-651X
• Publisher: The American Physical Society
• Country of Publication: United States
• Language: English

Related Subjects

71 CLASSICAL AND QUANTUM MECHANICS
GENERAL PHYSICS
ALGORITHMS
ANNEALING
LEARNING
ORIGIN
THERMAL EQUILIBRIUM