Fast Learning in an Actor-Critic Architecture with Reward and Punishment

A reinforcement architecture is introduced that consists of three complementary learning systems with different generalization abilities. The ACTOR learns state-action associations, the CRITIC learns a goal-gradient, and the PUNISH system learns what actions to avoid. The architecture is compared to the standard actor-crititc and Q-learning models on a number of maze learning tasks. The novel architecture is shown to be superior on all the test mazes. Moreover, it shows how it is possible to combine several learning systems with different properties in a coherent reinforcement learning framework.