Sebastien Le Digabel (Polytechnique Montreal)
Title :Â Blackbox optimization with the MADS algorithm and the NOMAD software
Abstract:Â NOMAD is a free software package for blackbox optimization. It implements the mesh adaptive direct search (MADS), a derivative-free optimization algorithm. NOMAD is designed to solve industrial problems, and therefore it includes several practical features such as constraints handling, the possibility of processing several objectives, noisy evaluations, etc. This seminar introduces derivative-free and blackbox optimization, the MADS algorithm and the practical use of NOMAD. Realistic test cases will also be discussed, with emphasis on the optimization of the hyperparameters of neural networks, for which NOMAD is naturally adapted thanks to its ability to consider categorical variables.
Zoom -Please contact : damien.tageddine [at] mail.mcgill.ca