Substructural local search in discrete estimation of distribution algorithms

Lima, Cláudio Miguel Faleiro de

http://hdl.handle.net/10400.1/471

Use this identifier to reference this record.

Name:	Description:	Size:	Format:
ClaudioLima-PhDThesis-2009.pdf		2.22 MB	Adobe PDF	Download

Send Feedback

Authors

Lima, Cláudio Miguel Faleiro de

Advisor(s)

Lobo, Fernando

Abstract(s)

The last decade has seen the rise and consolidation of a new trend of stochastic optimizers known as estimation of distribution algorithms (EDAs). In essence, EDAs build probabilistic models of promising solutions and sample from the corresponding probability distributions to obtain new solutions. This approach has brought a new view to evolutionary computation because, while solving a given problem with an EDA, the user has access to a set of models that reveal probabilistic dependencies between variables, an important source of information about the problem. This dissertation proposes the integration of substructural local search (SLS) in EDAs to speedup the convergence to optimal solutions. Substructural neighborhoods are de ned by the structure of the probabilistic models used in EDAs, generating adaptive neighborhoods capable of automatic discovery and exploitation of problem regularities. Speci cally, the thesis focuses on the extended compact genetic algorithm and the Bayesian optimization algorithm. The utility of SLS in EDAs is investigated for a number of boundedly di cult problems with modularity, overlapping, and hierarchy, while considering important aspects such as scaling and noise. The results show that SLS can substantially reduce the number of function evaluations required to solve some of these problems. More importantly, the speedups obtained can scale up to the square root of the problem size O( p `).