Laboratory of Parallel, Embedded architectures and Intensive Computing

L A P E C I

Multi-objective Optimization

Multi-objective Optimization (OPMO) team belongs to the laboratory of parallel and embedded architectures, and high performance computing (LAPECI) founded in April 2013. The team's research field is solving multi-objective optimization problems. These problems occur in various fields such as telecommunications, biology, physics, chemistry, finance, embedded systems, etc. These problems are usually multi-objective and for which we have to search solutions which optimize several conflicting objectives. In the field of embedded systems, for example, a multi-objective optimization problem which one faces is the problem of task mapping on a NoC (Network on Chip) where one must find a mapping which minimizes both the execution time and the energy consumption.
Solving such problems involves finding what is called Pareto optimal solutions. To find such solutions, two classes of methods can be used:

  • Exact methods such as branch-and-bound, dynamic programming, constraint programming, etc. Exact methods provide exact Pareto optimal solutions but are not useful to solve large size optimization problems and/or high number of objectives (> 2); in that case, processing time to solve these problems are prohibitive.
  • Approximate methods or metaheuristics such as evolutionary algorithms, particle swarm optimization, etc. Metaheuristics provide approximate solutions in reasonable delays.Optimization methods must present a good compromise between execution time and quality of solutions. This generally requires the hybridization of solving methods that have complementary features. Hybridization may involve the cooperation between metaheuristics only (e. g. cooperation of an evolutionary algorithm with a local search method) or between metaheuristics and exact methods. The resulting hybrid method generally requires huge computation time whose implementation needs the use of high performance computing techniques.
The main research theme of OPMO team is to provide efficient and effective optimization methods that use hybridization techniques and parallelism to solve large scale multi-objective optimization problems encountered in different areas of life.

+ Permanent members

+ PhD students

+ PhD theses

+ Publications

+ International conferences