A unified framework for credit assignment

摘要: Credit assignment is a central issue in variety of connectionist paradigms including artificial neural networks and genetic classifier systems. For networks, credit implicit the training algorithms used to update connection weights. In systems, it estimate rule strengths. This paper introduces unifying framework that covers helps explain differences similarities between large number approaches assignment, Q-learning, policy iteration, bucket brigade profit sharing. Each these can be formulated as version Adaptive Heuristic Critic (AHC). 11 refs.