Skip to content
Licensed Unlicensed Requires Authentication Published by De Gruyter 2020

3. Ant colony optimization and reinforcement learning

J. Michael Herrmann, Adam Price and Thomas Joyce

Abstract

Ant Colony Optimization (ACO) used to be one of the most frequently used algorithms in metaheuristic optimization. It can be interpreted as a Bayesian accumulation of information about an optimization problem, and bears some similarity to reinforcement learning. With this parentage, given a better theoretical understanding, ACO could be advanced into a practically competitive optimization algorithm. We are trying to support this development by studying the complex effects of parameters in ACO. For a toy example, which is, however, highly instructive in regard to the ability of the algorithm to overcome deception, we investigate how the parameter such as the number of ants and the rate of pheromone evaporation influences the performance. In addition, we investigate the suggestive link between ACO and reinforcement learning (RL) which does not only have the potential to facilitate the analysis of ACO, but which provides also options for the generation of new algorithms by the transfer of innovations across categories of algorithms. Based on these ideas, we propose an embedding of both ACO and RL into an joint evolutionary framework.

© 2020 Walter de Gruyter GmbH, Berlin/Munich/Boston