Q-PSP learning: an exploitation-oriented Q-learning algorithm and its applications

Proposes the Q-PSP learning method, which incorporates the the idea of the profit-sharing plan (PSP), as used in classifier systems, into Q-learning as a type of “exploitation-oriented” reinforcement learning in order to take advantage of the merits of these two approaches. By applying Q-PSP learning to several problems, it is shown that not only can a speed-up in learning be expected, but also the effectiveness for complex problems can be improved, and an appropriate balance between exploration and exploitation can be attained