An investigation of the use of local search in NP-hard problems

We combine local search algorithms with genetic algorithms. In this context local search can be thought of as learning over an individual's lifetime. We investigate two different ways of incorporating learning into the hybrid algorithm: Lamarckian evolution and the Baldwin effect. For each model we systematically vary the proportion of the population undergoing learning. We found that the quality of solution improves significantly at or above a critical level of learning.