IEEE CIS ACT Chapter Seminar: Network visualisations of multi-objective optimisation landscapes

#Computationally #expensive #optimisation; #Data-driven #Multi-objective #network #visualisation

IEEE Computational Intelligence Society, Australian Capital Territory (ACT) chapter invites you to: 

Virtual Seminar - 28 September 2021, 5-6pm AEST

Speaker: Professor Jonathan Fieldsend, Department of Computer Science, University of Exeter, UK

Title: Network visualisations of multi-objective optimisation landscapes

===

Abstract:  Local optima networks (LONs) have been developed over the last two decades to represent the landscape of (single objective) optimisation problems. In a LON, graph vertices represent local optima in the search domain, their radii the basin sizes, and directed edges between vertices the ability to transit from one basin to another (with the edge width denoting how easy this is). Recently, analogues of these network visualisations have been developed to visualise multi-objective landscapes. In this talk we discuss three different approaches for representing multi-objective landscapes with networks that have recently been developed: (i) the PLOS-net (Pareto Optimal Solutions Network), which uses an undirected graph, representing mutually non-dominating solutions and neighbouring links, but not basin sizes; (ii) the DNON (Dominance Neutral Optima Network) which utilises point-based Pareto hill-climbing to determine dominance neutral optima to construct a directed graph; and (iii) the PLON (Pareto Local Optima Network), which utilises set-based (Pareto local optima) representation to define mode construction. These alternative formulations will be compared on some illustrative problems, and some of the underlying computational issues in constructing LONs in a multi-objective as opposed to uni-objective problem domain will be discussed. The inherent issue of local dominance neutrality is also seen to be particularly important in such visualisations, as each a vertex in DNON and PLON constructs is typically a set. We also illustrate how using combinations of the alternative network visualisations can often grant additional insight to a problem. 

===

Speaker biography: Jonathan Fieldsend is Professor of Computational Intelligence, Director of Research, and Academic Lead of the Optimisation Group in the Department of Computer Science at the University of Exeter, UK. He graduated with a BA in Economics from the Durham University in 1998, an MSc in Computational Intelligence from the University of Plymouth in 1999 and a PhD in Computer Science from the University of Exeter in 2003. He has published widely in the technical literature, mainly on multi-objective optimisation, and its interface with machine learning. He has an h-index of 24 and i100-index of 10. His work has received £8M in grant funding spanning UKRI, industry and charities, including over £1M as Principal Investigator. He is an Associate Editor/Editorial Board Member of IEEE Transactions on Evolutionary Computation, ACM Transactions on Evolutionary Learning and Optimization, and Complex and Intelligent Systems. He is vice-Chair of the IEEE Computational Intelligence Society Task Force on Data-Driven Evolutionary Optimization of Expensive Problems and also vice-Chair of the IEEE Computational Intelligence Society Task Force on Multi-Modal Optimization. He was co-Chair of the Evolutionary Multi-criterion Optimization (EMO) Track at GECCO 2019 and GECCO 2020, and is Editor-in-Chief of GECCO 2022. 

============