Distributed Autonomous Design and Control of Evolutionary Algorithms

Location : LERIA, University of Angers, France

Salary : 1 931,17 € (net income per month)

Duration : 1 year Post Doc contract

Starting date : no later than May 2010

Supervision : Frédéric Saubion (U. Angers, France)

Abstract :

Evolutionary algorithms have been efficiently used for solving combinatorial problems. However, in order to obtain good performances, it is necessary to properly design the operators and to adjust their associated parameters. As for most metaheuristics methods, the performance of an evolutionary algorithm is intrinsically related to its ability to manage the balance between the exploitation and the exploration of the search space. Recently, new approaches have emerged to provide more autonomous algorithms, especially by automating the tuning and/or control of parameters. The purpose of this project is to define a new approach whose objective is twofold: on the one hand we want to control dynamically the behavior of operators in an evolutionary algorithm and, on the other hand, we want to manage a large set of potential operators, whose performances are a priori unknown. A mechanism identifies the best operators in order to reward them by increasing their application rate. We have already developed a controller that manages a set of operators and controls their use during solving. Nevertheless, learning and solving are combined into the resolution process, which may penalize the overall efficiency. Therefore, we propose to separate the different control agent into a distributed architecture in order to improve information extraction, learning and solving. This approach aims to reduce the user's tasks in the algorithm's design and parameter control, that could require a deep expertise or a tedious experimental process.

More information here

If interested, please contact :

Frédéric Saubion (Frederic.Saubion ‘at’ univ-angers.fr)

www.info.univ-angers.fr/pub/saubion

Send a full CV (including academic profile)