Automatically Configuring Multi-objective Local Search Using Multi-objective Optimisation

Abstract

Abstract. Automatic algorithm configuration (AAC) is becoming an increasingly crucial component in the design of high-performance solvers for many challenging combinatorial optimisation problems. This raises the question how to most effectively leverage AAC in the context of building or optimising multi-objective optimisation algorithms, and specifically, multi-objective local search procedures. Because the performance of multi-objective optimisation algorithms cannot be fully characterised by a single performance indicator, we believe that AAC for multi-objective local search should make use of multi-objective configuration procedures. We test this belief by using MO-ParamILS to automatically configure a highly parametric iterated local search framework for the classical and widely studied bi-objective permutation flowshop problem. To the best of our knowledge, this is the first time a multi-objective optimisation algorithm is automatically configured in a multi-objective fashion, and our results demonstrate that this approach can produce very good results as well as interesting insights into the efficacy of various strategies and components of a flexible multi-objective local search framework.

Automatic Configuration of Multi-Objective Local Search Algorithms for Permutation Problems

“…Our work presented here builds on and extends two recent, more limited studies (Blot, Pernet et al, 2017;. Here, we additionally consider the bi-objective TSP, to investigate to which extent our results for the PFSP generalise to another permutation-based MOP with uncorrelated objectives.…”

“…In our experiments, we will compare two SO-AAC approaches and one MO-AAC approach optimising the performance of a multi-objective local search algorithm. Specifically, as in previous work, we consider three distinct AAC approaches (Blot, Pernet et al, 2017):…”

“…However, the performance of MOO algorithms is generally assessed using multiple performance indicators, in order to characterise complementary properties of the solutions produced by them, such as convergence and diversity (Zitzler et al, 2003). This suggests the use of multi-objective configuration procedures that can effectively explore the trade-off between multiple performance indicators-the use of multi-objective configurators to optimise the performance of MOO algorithms (Blot, Pernet et al, 2017).…”

Automatic design of multi-objective local search algorithms

“…Recently, multi-objective con gurators such as SPRINT-race [23] or MOParamILS [2] have shown the possibility and the bene ts of using multiple performance indicators. It has also been shown that, on multi-objective AAC scenario optimising multiple multi-objective performance indicators of multi-objective target algorithms, using a multi-objective con gurator should be preferred to using a singleobjective con gurator with an aggregation of the performance indicators [3].…”

Algoritmo Híbrido para o Problema Flow Shop de Permutação Multiobjetivo

“…[Arroyo and Armentano 2005] propõem uma abordagem com busca local para otimizar dois pares de objetivos: (i) makespan e maximum tardiness; e (ii) makespan e total tardiness. A proposta em [Blot et al 2017]é o uso de ILS (Iterated Local Search) para otimizar makespan e total flowtime. Em [Ishibuchi et al 2003] os autores usam uma estratégia hill climbing simples em duas frentes: primeiramente são testados dois objetivos (makespan e maximum tardiness) e, num segundo teste o FSPé considerado com três objetivos (adicionando total flowtime).…”