• Login
• Admin

Burnett, Andrew (2021) Automated Heuristic Generation By Intelligent Search. PhD thesis, University of Nottingham.

Preview

PDF (Thesis - as examined) - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader Available under Licence Creative Commons Attribution. Download (45MB) | Preview

Abstract

This thesis presents research that examines the effectiveness of several different program synthesis techniques when used to automate the creation of heuristics for a local search-based Boolean Satisfiability solver.

Previous research focused on the automated creation of heuristics has almost exclusively relied on evolutionary computation techniques such as genetic programming to achieve its goal. In wider program synthesis research, there are many other techniques which can automate the creation of programs. However, little effort has been expended on utilising these alternate techniques in automated heuristic creation.

In this thesis we analyse how three different program synthesis techniques perform when used to automatically create heuristics for our problem domain. These are genetic programming, exhaustive enumeration and a new technique called local search program synthesis. We show how genetic programming can create effective heuristics for our domain. By generating millions of heuristics, we demonstrate how exhaustive enumeration can create small, easily understandable and effective heuristics. Through an analysis of the memoized results from the exhaustive enumeration experiments, we then describe local search program synthesis, a program synthesis technique based on the minimum tree edit distance metric. Using the memoized results, we simulate local search program synthesis on our domain, and present evidence that suggests it is a viable technique for automatically creating heuristics.

We then define the necessary algorithms required to use local search program synthesis without any reliance on memoized data. Through experimentation, we show how local search program synthesis can be used to create effective heuristics for our domain. We then identify examples of heuristics created that are of higher quality than those produced from other program synthesis methods. At certain points in this thesis, we perform a more detailed analysis on some of the heuristics created. Through this analysis, we show that, on certain problem instances, several of the heuristics have better performance than some state-of-the-art, hand-crafted heuristics.

Item Type:	Thesis (University of Nottingham only) (PhD)
Supervisors:	Parkes, Andrew Landa-Silva, Dario
Keywords:	heuristics, computational intelligence, algorithms
Subjects:	Q Science > Q Science (General) Q Science > QA Mathematics > QA 75 Electronic computers. Computer science T Technology > T Technology (General)
Faculties/Schools:	UK Campuses > Faculty of Science > School of Computer Science
Item ID:	74496
Depositing User:	Burnett, Andrew
Date Deposited:	05 Feb 2024 14:57
Last Modified:	05 Feb 2024 14:57
URI:	https://eprints.nottingham.ac.uk/id/eprint/74496

Actions (Archive Staff Only)

Edit View