BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Diffusion-based Bayesian Experimental Design - Jacopo Iollo (INRIA ) DTSTART:20250625T081500Z DTEND:20250625T091500Z UID:TALK232234@talks.cam.ac.uk DESCRIPTION:Bayesian Experimental Design (BED) is a powerful tool to reduc e the cost of running a sequence of experiments. When based on the Expecte d Information Gain (EIG)\, design optimization corresponds to the maximiza tion of some intractable expected contrast between prior and posterior dis tributions. Scaling this maximization to high-dimensional and complex sett ings where no closed form prior is available has been an issue due to BED inherent computational complexity. \;\nIn this work\, we introduce a p ooled posterior distribution with cost-effective sampling properties and p rovide a tractable access to the EIG maximization via a new gradient expre ssion. Diffusion-based samplers are used to compute the dynamics of the po oled posterior\, and ideas from bi-level optimization are leveraged to der ive an efficient joint sampling-optimization loop. The resulting efficienc y leverage the well-tested generative capabilities of diffusion models to BED to scenarios that were previously impractical. \;\nNumerical exper iments and comparison with state-of-the-art methods show the potential of the approach. As a practical application\, we showcase how our method acce lerates Magnetic Resonance Imaging (MRI) acquisition times while preservin g image quality. \;\nThis presentation will also detail how Diffuse\, a new modulable Python package for diffusion models\, facilitates composab ility and research in diffusion models through its simple and intuitive AP I\, allowing researchers to easily integrate and experiment with various m odel components.\nPresentation based on: Bayesian Experimental Design via Contrastive Diffusions. Iollo\, J.\, Heinkelé\;\, C.\, Alliez\, P.\,  \; Forbes\, F. (2025). International Conference on Learning Represent ations LOCATION:Seminar Room 1\, Newton Institute END:VEVENT END:VCALENDAR