SonOpt: Understanding the behaviour of bi-objective population-based optimisation algorithms through sound

SonOpt is an open source application that generates sound to facilitate a better understanding of the algorithmic behaviour of bi-objective population-based optimisation algorithms. At each generation of an algorithmic search process, SonOpt receives the current Pareto front approximation (the approximation set) and the hypervolume contributions of the approximation set points. It then produces three distinctive sound streams encapsulating information about the evolving shape of the approximation set, recurrence of approximation set points across generations and their location within the approximation set, and the distribution of hypervolume contributions within the approximation set. In turn, this information provides insights about convergence/stagnation of an algorithm, diversity in the approximation set, and relative importance of approximation set points. In practice, SonOpt is used alongside visualisation methods to facilitate multi-modal monitoring of the algorithmic search process. We demonstrate SonOpt's responsiveness via a numerical and audio analysis performed on a range of bi-objective optimisation problems using several well-known optimisation algorithms (NSGA-II, MOEA/D, multi-objective random search). SonOpt is available for download at https://github.com/tasos-a/SonOpt-2.0, and a range of videos is available at https://tinyurl.com/sonopt2 for live demonstrations of SonOpt. This is an extended abstract of [1].