|
Abstract : |
Shaping is an ancient animal training procedure that has also been studied by experimental psychologists. The principle underlying shaping is that learning to solve complex problems can be facilitated by first learning to solve related simpler problems. Although defining related simpler problems might be difficult when cognitive behavior such as language or mathematics is to be learned, it is relatively easy to determine a sequence of approximations that will lead to mastery of a target physical behavior. Indeed, shaping has been most useful for teaching motor skills to animals, and, for the same reason, shaping can also prove useful for training artificial learning systems to perform as controllers. In this paper, we present experimental results illustrating the utility of shaping in training controllers via reinforcement learning methods., |
