A hybrid learning strategy for discovery of policies of action

A hybrid learning strategy for discovery of policies of action

Ribeiro, Richardson and Enembreck, Fabrício and Koerich, Alessandro L.

Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 2006

Abstract : This paper presents a novel hybrid learning method and performance evaluation methodology for adaptive autonomous agents. Measuring the performance of a learning agent is not a trivial task and generally requires long simulations as well as knowledge about the domain. A generic evaluation methodology has been developed to precisely evaluate the performance of policy estimation techniques. This methodology has been integrated into a hybrid learning algorithm which aim is to decrease the learning time and the amount of errors of an adaptive agent. The hybrid learning method namely K-learning, integrates the Q-learning and K Nearest-Neighbors algorithm. Experiments show that the K-learning algorithm surpasses the Q-learning algorithm in terms of convergence speed to a good policy. © Springer-Verlag Berlin Heidelberg 2006.