A three-stage evolutionary method, the BBF-GA is introduced. BBF-GA is an acronym for building block filtering genetic algorithm. During the first stage, an ensemble of fast evolutionary algorithms is used to explore the search space. The best individual found by each of these evolutionary algorithms is propagated to the next phase. During the second stage, building block filtering is used to extract the essential parts of each of these local optimal strings, and masks these essential parts. During the third stage, a single evolutionary algorithm is used to find the global optimum by recombining the masked strings. For this purpose we use a special recombination operator that exploits the information stored in the masks. Given an appropriate basis, such that partial solutions can be discovered and evaluated in parallel and be combined afterwards, a recombination-based evolutionary algorithm can be very efficient. Therefore, learning of the structure of problem-spaces is important to make a more efficient recombination possible. The BBF-GA is a first step along this line for binary search spaces and problems that adhere to the building block hypothesis.

Optimization (acm G.1.6), Problem Solving, Control Methods, and Search (acm I.2.8)
Learning and adaptive systems (msc 68T05), Problem solving (heuristics, search strategies, etc.) (msc 68T20)
CWI
Software Engineering [SEN]

van Kemenade, C.H.M. (1998). Building block filtering and mixing. Software Engineering [SEN]. CWI.