On discretizing estimator-based learning algorithms
The authors illustrate the improvements gained by rendering various estimator algorithms discrete. Experimental results indicate that discretizing improves the performance of estimator algorithms. It is believed that discrete estimator algorithms (DEAs) constitute the fastest converging and most accurate learning automata reported to date. The DEAs are shown to have the monotone and moderation properties. Finally, any discretized learning automaton with monotone and moderation properties is proven to be ε-optimal in all stationary environments.
|Conference||Conference Proceedings of the 1991 IEEE International Conference on Systems, Man, and Cybernetics|
Lanctot, J.Kevin (J. Kevin), & Oommen, J. (1991). On discretizing estimator-based learning algorithms. In Proceedings of the IEEE International Conference on Systems, Man and Cybernetics (pp. 1417–1422).