Stochastic Local Search – Methods, Models, Applications

Hoos, H.H.
Pub. date
January 1999
215 of Dissertations in Artificial Intelligence
ISBN print
Artificial Intelligence, Computer & Communication Sciences, Computer Science

To date, stochastic local search (SLS) algorithms are among the standard methods for solving hard combinatorial problems from various areas of Artificial Intelligence and Operations Research. Some of the most successful and powerful algorithms for prominent problems like SAT, CSP, or TSP are based on stochastic local search.

This work investigates various aspects of SLS algorithms; in particular, it focusses on modelling these algorithms, empirically evaluating their performance, characterising and improving their behaviour, and understanding the factors which influence their efficiency. These issues are studied for the SAT problem in propositional logic as a primary application domain. SAT has the advantage of being conceptually very simple, which facilitates the design, implementation, and presentation of algorithms as well as their analysis. However, most of the methodology generalises easily to other combinatorial problems like CSP.

This Ph.D. thesis won the Best Dissertation Award 1999 (Dissertationspreis) of the German Informatics Society (Gesellschaft für Informatik).