An Exploration of Learning in Computational Analogy-making By

iAbstract The making of analogies is a powerful and potentially creative process that is both subtle and pervasive in human thought. Many cognitive processes, from the perception of a new situation to the design of a house, depend on analogical reasoning to produce an abstract mapping between two cases or domains. Computational models of analogy have been developed and used to solve problems by induction, but the systems are pattern-matching automatons lacking the flexibility and ingenuity of analogical reasoning in humans. Analogy is a situated act integrated into the cognitive processes of a constructive entity. This research investigates the role of an agent’s past experiences in computational analogy making. Learning and the biasing of perception and behaviour towards past experiences is a component of situated systems. A model of situated analogy is presented and its role in situated agents is described with particular attention to the development of situated analogical design agents. A learning module is integrated into an existing analogical problem-solving system using a reinforcement learning algorithm to develop a model of the problem-solver’s processes and behaviours. The system takes actions to affect the operation of the analogical system, perturbing it