Searching for general metaheuristics for optimization problems and knowledge management

Understanding the interaction between search algorithms and search domains is a very important problem in computer science. This interaction information can be referred to as metaknowledge, and can be stored using knowledge container. There is an interesting question about how to obtain such metaknowledge. Experiments have been used to acquire this metaknowledge for different search domains and algorithms. The experiments have been also conducted on some general combinatorial domains and the Brain-Computer Interface (BCI) domain.;The main accomplishments of this research are determining in an empirical way metaknowledge about heuristic search algorithms and their domains, designing the Adaptive Memory Programming (AMP) framework with unified parameters (Knowledge Container), incorporating the AMP algorithm into the BCI optimizer and obtaining very interesting results for the BCI optimizer.;The major part of the research deals with the idea of generalizing the search algorithm using the AMP framework. Using a conversion of parameters of other algorithms, there should be possible reuse of metaknowledge in other search algorithms and domains. The AMP algorithm framework is designed in order to answer the problem of metaknowledge reuse, and this objective is succeeded by unifying multiple heuristic search algorithms. Also the AMP implementation proves its usefulness in the BCI domain by finding very good solutions, which are much better than any previously known solutions.;Metaknowledge acquired during empirical research contains strong evidence of a logarithmic dependency between an average basin size in the search domain and the tabu search algorithm performance in comparison to other search algorithms. Thus metaknowledge was obtained in a very strict empirical way, showing importance of such experiments. It is possible to guide the metasearch algorithm (e.g., AMP) using acquired metaknowledge with an on-line evaluation of instance parameters in order to achieve the best search performance.