Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.596901
Title: Machine learning and automated theorem proving
Author: Bridge, J. P.
Awarding Body: University of Cambridge
Current Institution: University of Cambridge
Date of Award: 2010
Availability of Full Text:
Full text unavailable from EThOS.
Please contact the current institution’s library for further details.
Abstract:
Computer programs to find formal proofs of theorems were originally designed as tools for mathematicians, but modern applications are much more diverse. In particular they are used in formal methods to verify software and hardware designs to prevent errors being introduced into systems. Despite this, the high level of human expertise required in their use means that theorem proving tools are not widely used by non-specialists. The work described in this dissertation addresses one aspect of this problem, that of heuristic selection. In theory theorem provers should be automatic; in practice the heuristics used in the proof search are not universally optimal for all problems so human expertise is required to determie heuristic choice and to set parameter values. Modern machine learning has been applied to the automation of heuristic selection in a first order logic theorem prover. One objective was to find if there are any features of a proof problem that are both easy to measure and provide useful information for determining heuristic choice. Another was to determine and demonstrate a practical approach to making theorem provers truly automatic. In the experimental work, heuristic selection based on features of the conjecture to be proved and the associated axioms is shown to do better than any single heuristic. Additionally a comparison has been made between static features, measured prior to the proof search process, and dynamic features that measure changes arising in the early stages of proof search. Further work was done on determining which features are important, demonstrating that good results are obtained with only a few features required.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.596901  DOI: Not available
Share: