BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Youssef Diouane (remote) (Polytechnique Montréal) DTSTART:20260224T233000Z DTEND:20260225T003000Z DTSTAMP:20260513T193643Z UID:SFUOR/67 DESCRIPTION:Title: D irect-Search for Min-Max Derivative-Free Optimization\nby Youssef Diou ane (remote) (Polytechnique Montréal) as part of PIMS-CORDS SFU Operation s Research Seminar\n\nLecture held in ASB 10908.\n\nAbstract\nRecent appli cations in machine learning have renewed the community’s interest in min -max\noptimization problems. While gradient-based optimization methods are widely used to solve these\nproblems\, there exist many scenarios where s uch techniques are not well suited\, or even not applicable\,\nparticularl y when gradients are not accessible. In this talk\, we will investigate th e use of direct-search methods\, which belong to a class of derivative-fre e techniques that only require access to the objective function through an oracle. We will present a novel direct-search method for min-max saddle-p oint problems\, where the min and max players are updated sequentially. Th e convergence of this algorithm will be discussed in both deterministic an d stochastic settings. Finally\, experimental results related to robust op timization and Generative Adversarial Networks will be presented to illust rate how the proposed method can outperform commonly used optimization sch emes.\n LOCATION:https://researchseminars.org/talk/SFUOR/67/ END:VEVENT END:VCALENDAR