ABSTRACT
Many cognitive and computational processes can be seen as classification problems. Obvious examples include object recognition, concept learning, decision processes, diagnosis, and predictions, to name only a few. Classification can be achieved by rule-based or instance-based approaches, which were often seen as rivaling theories of cognition. Furthermore, the domain theory is used to adapt attribute weights depending on the goal. This makes use of psychological findings about higher weights for causal attributes and the systematicity principle. Rule-Enhanced Similarity will be implemented and evaluated in a multi-agent-system (MAS), where it will predict agent actions from world-states. In most MAS there exist only partial or weak domain theories, and instance-based methods that rely only on externally observable world-states are unlikely to approximate decisions of an agent that uses domain knowledge.
