The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9905

Title:Semantic Information in Similarity-Based Transfer: Reimplementing the MAC/FAC Model
Date: 1999
Abstract:Similarity-based transfer is a combination of cognitive processes that together form a versatile problem-solving strategy used by agents.The goal of similarity-based transfer is to bring previous experience to bear on a novel problem in an appropriate fashion, by retrieving appropriate knowledge that is likely to be applicable to the current situation, and then mapping between the retrieved knowledge and the current situation so that inferences can be generated and possible courses of action planned. Evidence presented by Gentner et al suggests that it is plausible to hold that there are two distinct processes active in the process of similarity-based transfer: (i) an initial retrieval process that is sensitive to the surface-features of knowledge, and which acts as acomputationally inexpensive prefilter on the potentially large set of episodic knowledge stored in a memory-pool; and (ii) a structural-alignment process that measures the degree of structural overlap between two episodes and provides a measure of the likelihood that the retrieved knowledge provides a sound basis from which to make inferences about the current situation.

[Search These Pages] [DAI Home Page] [Comment]