BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Kasper Bågmark (Chalmers) DTSTART:20260211T121500Z DTEND:20260211T130000Z DTSTAMP:20260422T161048Z UID:gbgstats/96 DESCRIPTION:Title: High-dimensional Bayesian filtering through deep density approximation\nby Kasper Bågmark (Chalmers) as part of Gothenburg statistics seminar \n\nLecture held in MVL14.\n\nAbstract\nIn this work\, we benchmark two re cently developed deep density methods for nonlinear filtering. Starting fr om the Fokker--Planck equation with Bayes updates\, we model the filtering density of a discretely observed SDE. The two filters: the deep splitting filter and the deep BSDE filter\, are both based on Feynman--Kac formulas \, Euler--Maruyama discretizations and neural networks. The two methods ar e extended to logarithmic formulations providing sound and robust implemen tations in increasing state dimension. Comparing to the classical particle filters and ensemble Kalman filters\, we benchmark the methods on numerou s examples. In the low-dimensional examples the particle filters work well \, but when we scale up to a partially observed $100$-dimensional Lorenz-9 6 model the particle-based methods fail and the logarithmic deep density m ethod prevails. In terms of computational efficiency\, the deep density me thods reduce inference time by roughly two to five orders of magnitude rel ative to the particle-based filters.\n LOCATION:https://researchseminars.org/talk/gbgstats/96/ END:VEVENT END:VCALENDAR