PPSN workshop on Multiobjective Problem Solving from Nature

"The origins and benefits of multiple objectives"

9th September 2006, Reykjavik, Iceland

URL: http://dbkgroup.org/knowles/MPSN3/

Oganizers: Joshua Knowles, David Corne and Kalyanmoy Deb

[Important Dates] [Programme and Paper Downloads]

Description

This workshop will take a philosophical, but grounded, look at the virtues (and vices) of the evolutionary and multiobjective approach to problem solving. In particular, participants will reflect on why, how and when multiple objectives arise in problems or problem formulations. We will also consider whether it is meaningful to think of Nature as solving multiobjective problems, and if so, what the limits of this metaphor are.

Workshop Programme

AM 9.00-12.30
9.00 Introduction to the workshop
9.15 Speaker: Carlos Coello A Survey of Constraint-Handling Techniques Based on Evolutionary Multiobjective Optimization
9.45 Speaker: Carlos Fonseca Preference Articulation in MOEAs
10.15 Mini-discussion and coffee break
11.00Speaker: Joshua Knowles Bias, Proxies and Solution Selection in Multiobjective Optimization
11.30Speaker: Yaochu Jin Modeling Regularity in Multi-Objective Optimization
12.00Speaker: Amiram Moshaiov Multiobjective Cybernetics and the Concept-Based Approach
PM 14.00-17.30
14.00 Welcome back, summary of morning session and discussion
14.30 Speaker: Jonathan Fieldsend Multiobjective Supervised Learning
15.00 Speaker: Hisao Ishibuchi Multiobjective Association Rule Mining
15.30 Coffee break
16.00 Speaker: Stefan Bleuler Reducing Bloat in GP with Multiple Objectives
16.30 Speaker: Arjun Chandra Multiobjective Ensemble Construction, Learning and Evolution
17.00-17.30 Discussion and final summary

Background Context

One of the principal strengths of an evolutionary approach to problem solving is that it allows real, messy, ill-behaved, hard problems to be tackled in a relatively unrefined form, and, as a result, the solutions that are found often work well, in practice. Multiobjective evolutionary algorithms (MOEAs), in principle, take this flexibility in problem solving to an even higher level as they can handle multiple, incommensurable, and conflicting definitions of solution quality, instead of requiring a single, overarching objective function to be given. The potential benefits of being able to handle multiple objectives includes being able to explore the tradeoffs between specific, well-defined goals — a practice that is very important in many engineering applications. However, multiple objectives need not be inherent to a problem; they may arise or be invoked for a variety of reasons, including, inter alia, as a way of describing elements of a desirable solution to a problem that is fundamentally very difficult to pose, or to help explore a very rugged (or very flat) search landscape better, or to handle constraints, or to decompose a problem. Recent research is recognizing this and it is the aim of this workshop to reflect on the new ways in which multiple objectives are being used, and to try to further our understanding of the resulting benefits and pitfalls. Issues arising out of this reflection are likely to include what the Pareto optima really mean in these different formulations, how the "best" Pareto optima can be identified, and how this view of problem solving relates to adaptation in Nature.

Important Dates

Workshop: 9 September, 2006.