A Generalized Approach to Handling Parameter Interdependencies in Probabilistic Modeling and Reinforcement Learning Optimization Algorithms

This paper generalizes our research on parameter interdependencies in reinforcement learning algorithms for optimization problem solving. This generalization expands the work to a larger class of search algorithms that use explicit search statistics to form feasible solutions. Our results suggest that genetic algorithms can both enrich and benefit from probabilistic modeling, reinforcement learning, ant colony optimization or other similar algorithms using values to encode preferences for parameter assignments. The approach is shown to be effective on both the Asymmetric Traveling Salesman and the Quadratic Assignment Problems. Introduction There has been a recent upsurge of interest in a family of search algorithms that store past experience not only as the best solutions generated, but also abstract representations of the decision processes employed. This interest is provoked by a number of factors. First, since solution memory is always limited, search algorithms which store only b..