Theoretical analysis of lexicase selection in multi-objective optimization

Abstract

Lexicase selection is a parent selection mechanism originally introduced for genetic programming that has also been considered in the context of multi-objective optimization. This is the first theoretical runtime analysis of lexicase selection showing results for the bi-objective leading ones trailing zeroes benchmark problem. The lexicase selection operator is embedded into a simple hillclimbing algorithm and compared with different selection operators from the literature that are based on the classical dominance relationship. Strengths and weaknesses of the operators are demonstrated providing insights into their working principles. Results of experiments accompany the theoretical findings and point towards interesting questions for future research.