Goal assignment through reinforcement and task embedding

摘要

In this paper we describe the technique of task embedding under the context of symbolic reinforcement learning. In particular, we focus on the issue of goal assignment from novice users to computational learners under a limited form of interaction through symbolic reinforcing cues. Reinforcing cue could be as simple as an indication of good or bad, and is symbolic in nature. It conveys the value judgment of the users, and serves as the only means of interaction for goal assignment in our study.