Quality-Diversity Methods for the Modern Data Scientist

Abstract

Unlike gradient-based methods, evolutionary algorithms use populations and exploit randomness to find novel and performant solutions. Quality-Diversity algorithms have recently emerged as a distinct paradigm designed to cultivate populations of simultaneously high-performing yet behaviorally diverse solutions. These algorithms show considerable success in challenging fields such as robotics and reinforcement learning. Despite their proven effectiveness and growing popularity, Quality-Diversity algorithms remain relatively underrecognized and underutilized in the broader data science landscape. This review aims to bridge this gap by providing a comprehensive introduction to the Quality-Diversity paradigm, elucidating its underlying philosophy, and synthesizing illustrative case studies for a general machine learning audience.