From a greedy Bouncy Particle Sampler to transport accelerated simulated annealing

Abstract: In this work we explore the connection between sampling and optimization. Starting from the Bouncy Particle Sampler, we combine a continuous time simulated annealing formulation with additional mode-seeking behaviour by biasing the exploration towards modes of the objective function. However, trying to tune the new parameters to achieve faster cooling leads us to intead consider a curve of probability measures. The second part of the talk is thus dedicated to considering curve following in a space of probability measures. To conclude, we demonstrate a particle algorithm inspired by this curve following idea.

numerical analysisoptimization and control

Audience: researchers in the topic

CAM seminar

Series comments: Online streaming via zoom on exceptional cases if requested. Please contact the organizers at the latest Monday 11:45.