Linkage learning through probabilistic expression Original Research Article

Linkage, in the context of genetic algorithms, represents the ability of building blocks to bind tightly together and thus travel as one under the action of the crossover operator. The goal of learning linkage has been intricately tied with defeating many of the bogeymen of GAs – building block disruption, inadequate exploration, spurious correlation and any number of other perceived stumbling blocks. Recent studies have shown that linkage can be learned in some very simple problems by simultaneously evolving problem representations alongside their solutions. This paper extends the applicability of these approaches by tackling their primary nemesis, the race between allelic selection and linkage learning.