Bounds of optimal learning
Conference paper
Belavkin, R. 2009. Bounds of optimal learning. 2009 IEEE International Symposium on Adaptive Dynamic Programming and Reinforcement Learning. Nashville, TN, USA 30 Mar - 02 Apr 2009 IEEE. pp. 199-204 https://doi.org/10.1109/ADPRL.2009.4927545
Type | Conference paper |
---|---|
Title | Bounds of optimal learning |
Authors | Belavkin, R. |
Abstract | Learning is considered as a dynamic process described by a trajectory on a statistical manifold, and a topology is introduced defining trajectories continuous in information. The analysis generalises the application of Orlicz spaces in non-parametric information geometry to topological function spaces with asymmetric gauge functions (e.g. quasi-metric spaces defined in terms of KL divergence). Optimality conditions are formulated for dynamical constraints, and two main results are outlined: 1) Parametrisation of optimal learning trajectories from empirical constraints using generalised characteristic potentials; 2) A gradient theorem for the potentials defining optimal utility and information bounds of a learning system. These results not only generalise some known relations of statistical mechanics and variational methods in information theory, but also can be used for optimisation of the exploration-exploitation balance in online learning systems. |
Research Group | Artificial Intelligence group |
Conference | 2009 IEEE International Symposium on Adaptive Dynamic Programming and Reinforcement Learning |
Page range | 199-204 |
Proceedings Title | 2009 IEEE Symposium on Adaptive Dynamic Programming and Reinforcement Learning |
ISSN | 2325-1824 |
Electronic | 2325-1867 |
ISBN | |
Hardcover | 9781424427611 |
Publisher | IEEE |
Publication dates | |
Mar 2009 | |
15 May 2009 | |
Publication process dates | |
Deposited | 24 Mar 2010 |
Output status | Published |
Web address (URL) | http://www.eis.mdx.ac.uk/staffpages/rvb/publications/rvb-adprl09.pdf |
Digital Object Identifier (DOI) | https://doi.org/10.1109/ADPRL.2009.4927545 |
Scopus EID | 2-s2.0-67650458719 |
Web of Science identifier | WOS:000271705100028 |
Web address (URL) of conference proceedings | https://ieeexplore.ieee.org/xpl/conhome/4910084/proceeding |
Language | English |
File |
https://repository.mdx.ac.uk/item/8206y
Download files
37
total views12
total downloads1
views this month0
downloads this month