BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Sampling with kernelized Wasserstein gradient flows - Anna Korba ( ENSAE (École Nationale de la Statistique et de l'Administration)) DTSTART:20220429T080000Z DTEND:20220429T090000Z UID:TALK171896@talks.cam.ac.uk DESCRIPTION:Sampling from a probability distribution whose density is only known up to a normalisation constant is a fundamental problem in statisti cs and machine learning. Recently\, several algorithms based on interactiv e particle systems were proposed for this task\, as an alternative to Mark ov Chain Monte Carlo methods or Variational Inference. \;\nThese parti cle systems can be designed by adopting an optimisation point of view for the sampling problem: an optimisation objective is chosen (which typically measures the dissimilarity to the target distribution)\, and its \;Wa sserstein gradient flow is approximated by an interacting particle system.  \; At stationarity\, the stationarity states of these particle system s define an empirical measure approximating the target distribution. \ ;​\nIn this talk I will present recent work on such algorithms\, such as Stein Variational Gradient Descent [1] \;or Kernel Stein Discrepancy Descent [2]\, two algorithms based on Wasserstein gradient flows and repro ducing kernels. \; \;\nI will discuss some recent results\, that s how that these particle systems can provide a good approximation of the ta rget distribution\; as well as current issues and \;open questions on the empirical and theoretical side.\n[1] \; \;A non-asymptotic Ana lysis of \;Stein Variational Gradient Descent. Korba\, A.\, Salim\, A. \, Arbel\, M.\, Luise\, G.\, Gretton\, A. Neural Information Processing Sy stems (Neurips)\, 2020\n[2] Kernel Stein Discrepancy Descent. Korba\, A.\, Aubin-Frankowski\, P.C.\, Majewski\, S.\, Ablin\, P. International Confer ence of Machine Learning (ICML)\, 2021. LOCATION:Seminar Room 1\, Newton Institute END:VEVENT END:VCALENDAR