Visualising many-objective populations with treemaps

Abstract

Visualising populations of solutions is an important aspect of evolutionary computation (EC), allowing an algorithm user to evaluate the performance of an algorithm and a decision maker to understand the solution set from which they must choose an operating solution. We present a novel approach to visualising multi-objective data, employing treemaps to display both solutions and objectives. We define a simple approach to constructing a tree that can represent a multi-objective population in terms of dominance, and illustrate several ways in which it can be used. Examples are provided that reveal characteristics of objective space, as well as combining information about the parameter space component of the population. The paper concludes with a discussion about the further potential of treemaps within EC.