Thursday 20 Feb 2014: Visualising Many-objective Populations
Dr David Walker - Computer Science, University of Exeter
Harrison 103 15:00-16:00
Optimisation problems often comprise a large set of objectives, and visualising the set of solutions to a problem can help with understanding them, assisting a decision maker. If the set of objectives is larger than three, visualising solutions to the problem is a difficult task. Techniques for visualising high-dimensional data are often difficult to interpret. Conversely, discarding objectives so that the solutions can be visualised in two or three spatial dimensions
results in a loss of potentially important information. This talk demonstrates novel techniques for visualising many-objective populations; one of them uses the complete set of objectives to present solutions in a clear and intuitive fashion while the other compresses the objectives of a population into two dimensions while minimising the information that is lost. Methods for identifying landmark solutions in a non-dominated solution set are also demonstrated. All of the techniques are illustrated on populations of solutions to optimisation test problems.