This paper presents an overview of the field of Stochastic Learning Automata (LA), and concentrates, in particular, on the most recent advances in the field. LA, which have been studied for more than three decades, are probabilistic finite state machines that can model how biological systems can learn. The structure of such a machine can be fixed, or it can be changing with time. A LA can also be implemented by using action probability updating rules which may or may not depend on estimates from the Environment being investigated. While, traditionally, these updating rules have worked with the continuous probability space, we1 will explain how LA can be designed by discretizing the probability space. The paper2 will describe the design and analysis of both continuous and dis-cretized LA, and will highlight the subtle differences between the corresponding learning machines, their convergence properties, and their learning capabilities. The paper will then discuss the most recent developments such as the Generalized Thathachar-Sastry estimator scheme. The paper also includes a comprehensive list of the applications in which LA have proven their powerful potential.

Additional Metadata
Keywords Continuous learning automata, Discretized learning automata, Estimator-based learning automata, Learning Automata
Persistent URL dx.doi.org/10.1109/ICCET.2010.5485366
Conference 2010 2nd International Conference on Computer Engineering and Technology, ICCET 2010
Citation
Oommen, J. (2010). Recent advances in Learning Automata systems. In ICCET 2010 - 2010 International Conference on Computer Engineering and Technology, Proceedings. doi:10.1109/ICCET.2010.5485366