Section: Partnerships and Cooperations

International Initiatives

• Collaboration with University of Mons (UMONS). The collaboration consists mainly in the joint supervision of three Ph.D theses: the thesis of Rudi Leroy defended on November 19^th, 2015, the thesis of Jan Gmys started last year, and the thesis of Gautier Vaillant started in September 2015.

s3-bbo

• Title: Threefold Scalability in Any-objective Black-Box Optimization (s3-bbo)

• International Partner (Institution - Laboratory - Researcher):

  • Shinshu University, Japan

• Duration: 2015-2017

• See also: http://francejapan.gforge.inria.fr/doku.php?id=associateteam

• The main scientific goals of this collaboration is to theoretically derive, analyze, design, and develop scalable evolutionary and other stochastic local search algorithms for large-scale optimization considering three different axes of scalability: (i) decision space, (ii) objective space, and (iii) availability of distributed and parallel computing resources. This research will allow us to design, control, predict, analyze and optimize parameters of recent complex, large-scale, and computationally expensive systems, providing the basic support for problem solution and decision-making in a variety of real world applications. For single-objective continuous optimization, we want to theoretically derive variants of the state-of-the-art CMA-ES with linear time and space complexity scalings with respect to the number of variables. We will exploit the information geometry framework to derive updates using parametrization of the underlying family of probability distribution involving a linear number of components. The challenges are related to finding good representations that are theoretically tractable and meaningful. For the design of robust algorithms, implementing the derived updates, we plan to follow the same approach as for the design of CMA-ES. For multi- and many-objective optimization, we will start by characterizing and defining new metrics and methodologies to analyze scalability in the objective space and in terms of computational resources. The first challenge is to accurately measure the impact of adding objectives on the search behavior and on the performance of evolutionary multi- and many- objective optimization (EMyO) algorithms. The second challenge is to investigate the new opportunities offered by large-scale computing platforms to design new effective algorithms for EMyO optimization. To this end, we plan to follow a feature-based performance analysis of EMyO algorithms, to design new algorithms using decomposition-based approaches, and to investigate their mapping to a practical parallel and distributed setting.

• Memorandum of Understanding between Shinshu University (Nagano, Japan) and Inria, signed on March 2014.

• University of Coimbra, Portugal.

• IRIDIA, Université Libre de Bruxelles.

• Cologne University of Applied Sciences, Germany.

• Leiden University, Netherlands.

• UMONS University and Tractebel company, Belgium.

• EMI - Univ. Rabat, Morocco.

• Univ. Oviedo, Spain.

• Univ. Istanbul, Turkey.

• University of KULAK Team Codes (Belgium) - Data science for Optimization

• University of KENT (UK) - Knowledge and Optimization

• University of Aberdeen (UK) - Fitness landscape, representation and performance

• University of British Columbia, Canada

• University of Münster, Germany

• JSPS-MEXT project on Evolutionary multi-objective optimization, landscape analysis, and search performance, with Shinshu University, Nagano, Japan (2013—2016).

• Excellencia project(2015-2017) with University of Valencia (Spain) and University of Oviedo (Spain) on intelligent techniques for robust scheduling and energy-aware transportation systems.