Section: New Results

Interactive Evolutionary Algorithms for Visual decision making

Participants : Nadia Boukhelifa, Waldo Cancino, Jean-Daniel Fekete, Evelyne Lutton [correspondant] .

When dealing with very large datasets with many dimensions, it is often difficult to efficiently navigate and find interesting viewpoints, significative compound variables, unexpected behaviour, and other remarkable characteristics.

Our aim within the System@tic CSDL project (Complex Systems Design Lab, 2009–2012) is to use interative evolutionary algorithm to assist the user in its exploration task. Finding an interesting, non obvious, viewpoint on a complex dataset can be formulated as an interactive optimisation problem. Population-based evolutionary search mechanisms can then efficiently be exploited for suggesting new viewpoints on data, that progressively adapt to the needs of an user.

In September 2011 (arrival of Nadia Boukhelifa and Waldo Cancino) we started to build a prototype based on GraphDice, that proposes new dimensions in the scatterplot matrix. These secondary set of dimensions are compositions of the dimensions of the initial datased. Starting from an initial set of suggested dimensions (PCA analysis of the dataset), an evolutionary algorithm progressively refines the compound dimensions according to a measurement of the activity of the user on the corresponding views.