This is a past event
Abstract:
Recent advances in evolutionary optimisation algorithms mean that it is possible to find approximations to the optimal trade-off between several objectives. With only two or three objectives visualising the trade-off and relations between solutions is straightforward, but with more objectives understanding the solutions and available trade-offs is difficult. This talk will briefly review multi- and many-objective optimisation and present a number of new techniques for visualising and understanding the resulting set of mutually non-dominating solutions. The "dominance distance" a new metric for measuring the distance between non-dominated solutions will presented, which allows projections of the mutually non-dominated solution set into lower dimensions. The "edges" of solution sets are important landmarks and a number of ways of defining the edges of mutually non-dominating sets will be described. This leads to new insights into the structure and geometry of high-dimensional mutually non-dominating sets. These tools are also applied to the visualisation of league tables, which are examined from a multi-objective optimisation point of view.
Bio:
Richard Everson is Professor of Machine Learning and Head of Computer Science at Exeter University. His research interests lie in data analysis and modelling using statistical pattern recognition, and in multi-objective optimisation. He graduated from Cambridge University with a degree in Natural Sciences, concentrating on Physics, and did research on chaotic dynamical systems for a PhD at Leeds University. Interests in the analysis of complex data took him to Brown, Yale and the Rockefeller Universities in the USA to work on analysis fluid flows and brain imaging data. He returned briefly to Imperial College and has been in the Computer Science department at Exeter since 1999.
- Speaker
- Prof. Richard Everson, Computer Science, University of Exeter
- Hosted by
- Chenghua Lin
- Venue
- MT 203