This thesis is concerned with reasoning by analogy within the context of auto-mated problem solving. In particular, we consider the provision of an analogical reasoning component to a resolution theorem proving system. The framework for reasoning by analogy which we use (called Basic APS) contains three major components -the finding of analogies (analogy matching), the construction of analogical plans, and the application of the plans to guide the search of a theorem prover. We first discuss the relationship of analogy to other machine learning techniques. We then develop programs for each of the component processes of Basic APS. First we consider analogy matching. We reconstruct, analyse and crticise two previous analogy matchers. We introduce the notion of analogy heuristics in order to understand the matchers. We find that we can explain the short-comings of the matchers in terms of analogy heuristics. We then develop a new analogy matching algorithm, based on flexible application of analogy heuristics, and demonstrate its superiority to the previous matchers. We go on to consider analogical plan construction. We describe procedures for constructing a plan for the solution of a problem, given the solution of a different problem and an analogy match between the two problems. Again, we compare our procedures with corresponding ones from previous systems. We then describe procedures for the execution of analogical plans. We demon-strate the procedures on a number of example analogies. The analogies involved are straightforward for a human, but the problems themselves involve.huge search spaees, if tackled directly using resolution. By comparison with unguided search, we demonstrate the dramatic reductfon in search entaile_d by the use of an ana-logical plan. We then consider some directions for development of our analogy systems, which have not yet been implemented. Firstly, towards more flexible and power-ful execution of analogical plans. Secondly, towards an analogy system which can improve its own ability to find and apply analogies over the course of experience.
| Identifer | oai:union.ndltd.org:bl.uk/oai:ethos.bl.uk:384228 |
| Date | January 1988 |
| Creators | Owen, Stephen G. |
| Publisher | University of Edinburgh |
| Source Sets | Ethos UK |
| Detected Language | English |
| Type | Electronic Thesis or Dissertation |
| Source | http://hdl.handle.net/1842/27156 |
Page generated in 0.0019 seconds