Comparison of GA-Based Algorithms: A Viewpoint of Learning Scheme

Abstract

Learning is at the core of intelligence. How does learning work in nature-inspired optimization algorithms? This paper tries to answer this question by analyzing the learning mechanisms in Genetic Algorithm (GA) based Algorithms (GAAs). First, we give a learning scheme, which includes four basic elements including learning subject, learning object, learning result and learning rules. Different GAA has different learning mechanisms. Each GAA generates new solutions by learning to explore/exploit promising sub-space. The learning mechanism of three kinds of GAA, including GA, evolutionary strategy and differential evolution, are studied. We study the learning mechanism from the viewpoint of evolutionary operators, including selection, crossover (or recombination) and mutation. This study enables us to get more insights of GAAs.