Using fitness distributions to design more efficient evolutionary computations

There is a need for methods to generate more efficient and effective evolutionary algorithms. Traditional techniques that rely on schema processing, minimizing expected losses, and an emphasis on particular genetic operators have failed to provide robust optimization performance. An alternative technique for enhancing both the expected rate and probability of improvement in evolutionary algorithms is proposed. The method is investigated empirically and is shown to provide a potentially useful procedure for assessing the suitability of various variation operators in light of a particular representation, selection operator, and objective function