Reinforcement Learning with Pattern-based Rewards

J.F. Peters, C. Henry, and S. Ramanna (Canada)


Approximation space, ecosystem, intelligent systems, rein forcement learning, rough sets, swarm.


This paper introduces an approach to deriving pattern based rewards during reinforcement learning by cooper ating agents. Rough set theory introduced by Zdzisław Pawlak in 1982 provides a ground for deriving pattern based rewards in the context of approximation spaces. The framework provided by an approximation space makes it possible to derive pattern-based reference rewards used to compute action rewards as well as action preferences. Ap proximation spaces are used to derive action-based refer ence rewards at the swarm intelligence level. Two differ ent forms of reinforcement comparison are considered as a part of a study of learning in real-time by a swarm. In addition, this article introduces an artificial ecosystem test bed that makes it possible to study learning by collections of biologically-inspired bots. The contribution of this arti cle is the introduction of an approach to rewarding swarm behavior in the context of approximation spaces.

Important Links:

Go Back