The BONUS research group aims to develop new approaches capable of effectively tackling the massive and complex nature of modern optimization problems. Solving such problems poses several challenges, notably related to dimensionality, scalability, and the efficient use of parallelization on massively multi-core and heterogeneous supercomputers, incorporating other parallel devices such as GPU accelerators. The team is interested in addressing difficult optimization problems with large number of variables, multiple objectives, blackbox functions, expensive functions, multimodal functions, etc. It aims to develop and analyze optimization methods at the interface of computational intelligence (e.g., evolutionary algorithms, fitness landscape analysis), machine learning (e.g., surrogates, automated algorithm configuration/selection, autoML), and high-performance parallel computing (e.g., massively parallel and heterogeneous supercomputers).

More specifically, the research activities of BONUS are structured around three main axes: (i) decomposition-based optimization in the decision space and objective space, (ii) intelligent optimization and machine learning through the design and analysis of automated, model- and data-guided high-level advanced approaches, and (iii) ultra-scale optimization on massively parallel, distributed, and heterogeneous computing architectures. The first axis addresses challenges related to the dimensionality of optimization problems. The second axis tackles challenges related to the complexity and intrinsic structure of the search landscape induced by an optimization problem or instance. The third axis aims to efficiently exploit the increasing amount of available computing resources. The combination of these three axis is rarely addressed in the literature and constitutes a challenge in itself. From a software perspective, the team aims to systematically integrate its developments into software frameworks such as ParadisEO. In terms of applications and industrial transfer, the team focuses particularly on scheduling energy systems such as smart grids and engineering design.