Bounds of optimal learning.
Belavkin, Roman V. (2009) Bounds of optimal learning. In: 2009 IEEE International Symposium on Adaptive Dynamic Programming and Reinforcement Learning, March 30 – April 2, 2009, Sheraton Music City Hotel, Nashville, TN, USA.
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.
|Item Type:||Conference or Workshop Item (Paper)|
|Research Areas:||A. > School of Science and Technology > Computer Science
A. > School of Science and Technology > Computer Science > Artificial Intelligence group
|Depositing User:||Dr Roman Belavkin|
|Date Deposited:||24 Mar 2010 14:08|
|Last Modified:||06 Dec 2016 17:29|
Actions (login required)