Achieving Memetic Adaptability by Means of Agent-Based Machine Learning

Abstract

Over recent years, there has been increasing interest of the research community towards evolutionary algorithms, i.e., algorithms that exploit computational models of natural processes to solve complex optimization problems. In spite of their ability to explore promising regions of the search space, they present two major drawbacks: 1) they can take a relatively long time to locate the exact optimum and 2) may sometimes not find the optimum with sufficient precision. Memetic Algorithms are evolutionary algorithms inspired by both Darwinian principles and Dawkins' notion of a meme, able not only to converge to high-quality solutions, but also search more efficiently than their conventional evolutionary counterparts. However, memetic approaches are affected by several design issues related to the different choices that can be made to implement them. This paper introduces a multiagent-based memetic algorithm which executes in a parallel way different cooperating optimization strategies in order to solve a given problem's instance in an efficient way. The algorithm adaptation is performed by jointly exploiting a knowledge extraction process together with a decision making framework based on fuzzy methodologies. The effectiveness of our approach is tested in several experiments in which our results are compared with those obtained by nonadaptive memetic algorithms. The superiority of the proposed strategy is manifest in the majority of cases.