The primary computational bottleneck in molecular dynamics is the timescale gap between the microscopic timesteps required for numerical integration and the macroscopic timescales of biological processes. This talk demonstrates how generative models address this bottleneck across different step sizes. At the extreme of very large timescales, we show how to train diffusion models to efficiently sample from the equilibrium distribution. With proper parameterization, an energy-based diffusion model can also allow us to extract physical forces for simulation.<\/p>\n\n\n\n
However, practical simulations typically operate in the intermediate regime. To address this, we introduce Hamiltonian flow maps to achieve stable macroscopic steps. By leveraging the mean flow paradigm to predict evolution over a chosen time span, this framework enables updates far beyond the stability limits of classical integrators while faithfully recovering system dynamics. Crucially, it trains directly on independent, trajectory-free datasets, avoiding the cost of expensive future-state rollouts. Together, flow maps and diffusion models provide means to efficiently simulate molecular systems across all relevant timescales.<\/p>\n\n\n\n
Michael Plainer is a second-year Ph.D. student in computer science at the Freie Universit\u00e4t Berlin and the ELIZA Zuse School, co-supervised by Frank No\u00e9 and Klaus-Robert M\u00fcller. He obtained his BSc. and an MSc. at the Technical University of Munich, where he worked in the lab of Stephan G\u00fcnnemann.<\/p>\n\n\n\n
Winfried Ripken is a second-year Ph.D. student in computer science at the Technical University Berlin and the BIFOLD graduate school, co-supervised by Stefan Chmiela and Klaus-Robert M\u00fcller. Prior to the PhD, Winfried worked as Machine Learning Researcher at Merantix Momentum and obtained his MSc. in computer science from the Hasso Plattner Institute.<\/p>\n\n\n\n
Gregor Lied is a first-year master\u2019s student in computer science at the Technical University of Berlin, where he works with Michael Plainer and Winfried Ripken under the co-supervision of Stefan Chmiela and Klaus-Robert M\u00fcller.<\/p>\n","protected":false},"excerpt":{"rendered":"
The primary computational bottleneck in molecular dynamics is the timescale gap between the microscopic timesteps required for numerical integration and the macroscopic timescales of biological processes. This talk demonstrates how generative models address this bottleneck across different step sizes. At the extreme of very large timescales, we show how to train diffusion models to efficiently […]<\/p>\n","protected":false},"featured_media":1172625,"template":"","meta":{"msr-url-field":"","msr-podcast-episode":"","msrModifiedDate":"","msrModifiedDateEnabled":false,"ep_exclude_from_search":false,"_classifai_error":"","msr_hide_image_in_river":0,"footnotes":""},"research-area":[13556],"msr-video-type":[270340],"msr-locale":[268875],"msr-post-option":[],"msr-session-type":[],"msr-impact-theme":[],"msr-pillar":[],"msr-episode":[],"msr-research-theme":[],"class_list":["post-1172624","msr-video","type-msr-video","status-publish","has-post-thumbnail","hentry","msr-research-area-artificial-intelligence","msr-video-type-msr-new-england-generative-modeling-sampling-seminar","msr-locale-en_us"],"msr_download_urls":"","msr_external_url":"https:\/\/youtu.be\/8WYcWrHt-oY","msr_secondary_video_url":"","msr_video_file":"http:\/\/0","_links":{"self":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1172624","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video"}],"about":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/types\/msr-video"}],"version-history":[{"count":2,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1172624\/revisions"}],"predecessor-version":[{"id":1172685,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video\/1172624\/revisions\/1172685"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media\/1172625"}],"wp:attachment":[{"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/media?parent=1172624"}],"wp:term":[{"taxonomy":"msr-research-area","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/research-area?post=1172624"},{"taxonomy":"msr-video-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-video-type?post=1172624"},{"taxonomy":"msr-locale","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-locale?post=1172624"},{"taxonomy":"msr-post-option","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-post-option?post=1172624"},{"taxonomy":"msr-session-type","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-session-type?post=1172624"},{"taxonomy":"msr-impact-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-impact-theme?post=1172624"},{"taxonomy":"msr-pillar","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-pillar?post=1172624"},{"taxonomy":"msr-episode","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-episode?post=1172624"},{"taxonomy":"msr-research-theme","embeddable":true,"href":"https:\/\/www.microsoft.com\/en-us\/research\/wp-json\/wp\/v2\/msr-research-theme?post=1172624"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}