This session was part of the PGMO-COPI'14 conference that took place in October 2014 at Ecole Polytechnique.
Many numerical optimization problems can only be handled as black-box problems where no derivatives are available. This session focuses on a specific class of methods for the optimization of such problems namely randomized comparison-based or function-value-free (FVF) methods that use the objective function through comparisons exclusively, i.e., updates of the algorithm's state variables use comparisons between queried solutions only. Those methods are quite successful for the optimization of difficult problems where the typical difficulties are related to non-convexity, noise, multi-modality, non-separability, ill-conditionning. This session will introduce state-of-the-art randomized comparison-based methods in particular the Covariance-Matrix-Adaptation Evolution Strategy (CMA-ES) and related variants. It will feature recent advances for single and multi-objective optimization with a specific focus on large-scale optimization. Theoretical concepts for algorithm design related to invariances, information geometry and their application for constructing robust and efficient algorithms will be discussed. Practical aspects of the methods will be covered such that the session should be of particular interest for persons interested to apply the methods to their specific problem.