Summary

Heuristic search methods have been applied to a wide variety of optimisation problems. A central element of these algorithms ’ success is the correct choice of values for their control parameters. To tune these settings, the use of specialists ’ knowledge and experience are often required. In this thesis, we first formalise the problem of parameter adaptation in heuristic search. Thereafter, we propose an automated mechanism, i.e. a method that reduces the strong dependency on experts, for choosing the best performing algorithm among several heuristic search approaches and optimis-ing its parameters. The novel Multiple Algorithms ’ Parameter Adaptation Algorithm (MAPAA) is based on Population-Based Incremental Learning, a method that combines the concepts of Competitive Learning and Genetic Algorithms. Addressing specific characteristics of the adaptation scenario, we extend the basic approach to deal with more versatile search spaces and to run successfully for small populations and few generational cycles. All newly introduced techniques are analysed in detail, and the efficiency and robustness of the MAPAA is studied and proven in several applications. I Acknowledgements I would like to express my deepest gratitude to my supervisor Professor Edward Tsang. He has been a remarkable mentor. Only his guidance, en-couragement and seemingly infinite patience have made my research work possible