BEGIN:VCALENDAR VERSION:2.0 PRODID:researchseminars.org CALSCALE:GREGORIAN X-WR-CALNAME:researchseminars.org BEGIN:VEVENT SUMMARY:Petar Jovanovski (Chalmers University of Technology & University o f Gothenburg) DTSTART:20230420T111500Z DTEND:20230420T120000Z DTSTAMP:20260422T155153Z UID:gbgstats/23 DESCRIPTION:Title: Approximate Bayesian Computation with Backward Simulation for Discretely Observed Diffusions\nby Petar Jovanovski (Chalmers University of Tech nology & University of Gothenburg) as part of Gothenburg statistics semina r\n\nLecture held in MVL14.\n\nAbstract\nStochastic differential equations (SDE) are employed in many areas of science as a powerful tool for modell ing processes that are subject to random fluctuations. Bayesian inference for a large class of SDEs is challenging due to the analytic intractabilit y of the likelihood function. Nevertheless\, forward simulation via numeri cal methods is straightforward\, motivating the use of approximate Bayesia n computation (ABC). We propose a simulation scheme for SDE models that is based on processing the observation in both the forward and backward dire ction\, effectively utilizing the information provided by the observed dat a. This leads to the simulation of sample paths that are consistent with t he observations\, thereby increasing the ABC acceptance rate. We additiona lly leverage partial exchangeability of Markov processes and employ invari ant neural networks to learn the summary statistics that are needed in ABC . These are sequentially learned by exploiting a sequential Monte Carlo AB C sampler\, which provides new training data at each iteration. Therefore\ , our novel contribution is a learning tool for SDE model parameters while simultaneously learning the summary statistics. Using synthetic data gene rated from the Chan-Karaolyi-Longstaff-Sanders SDE family\, we show that o ur approach accelerates inference considerably\, compared to standard (for ward-only) methods\, while preserving inference accuracy.\n LOCATION:https://researchseminars.org/talk/gbgstats/23/ END:VEVENT END:VCALENDAR