BEGIN:VCALENDAR
VERSION:2.0
PRODID:OpenCms 20.0.18
BEGIN:VTIMEZONE
TZID:Europe/Berlin
X-LIC-LOCATION:Europe/Berlin
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700329T020000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=3
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701025T030000
RRULE:FREQ=YEARLY;BYDAY=-1SU;BYMONTH=10
END:STANDARD
END:VTIMEZONE				
BEGIN:VEVENT
DTSTAMP:20251005T165832
UID:c29fa191-a1fb-11f0-bb1e-000e0c3db68b
SUMMARY:ESPResSo Summer School: Dr. Simon Olsson
DESCRIPTION:Large time-step all-atom molecular dynamics with deep generative models\n \nWe introduce TITO, a transferable generative framework for molecular dynamics (MD) built on our prior developments in surrogate modeling of stochastic dynamics. Previously, we proposed Implicit Transfer Operator (ITO) Learning, a method based on denoising diffusion probabilistic models to learn transition operators across multiple time resolutions—enabling stable, self-consistent MD surrogates using timesteps up to six orders of magnitude longer than conventional MD and even coarse-grained representations of all-atom behavior [1,2].\nTITO extends ITO’s capabilities by learning transferable implicit transfer operators across molecular systems. It directly models the transition probability density p(x(Δt)∣x(0)) for arbitrarily long Δt, trained jointly across multiple timestep scales and multiple different systems. The surrogate preserves critical well-founded properties of molecular dynamics and allows explicit control over compute-accuracy trade-offs.\nTrained on small molecules and short peptides, TITO achieves up to a million-fold acceleration, facilitating quantitative simulation of equilibrium properties and non-equilibrium relaxation dynamics, as well as qualitative thermodynamic and kinetic predictions in larger peptides (twice the size encountered during training).\nOverall, TITO demonstrates that advanced deep generative models can substantially accelerate atomistic simulations without sacrificing physical fidelity and can generalize across chemical space—opening new opportunities for studying slow, experimentally relevant phenomena (e.g., folding, binding/unbinding) at unprecedented scales.\n \n[1] M. Schreiner, O. Winther, and S. Olsson, 'Advances in Neural Information Processing Systems', volume 36, Curran Associates, USA (2023)\n[2] J. Viguera Diez, M. Schreiner, O. Engkvist, and S. Olsson, arXiv preprint 2410.10605[physics.chem-ph], April 2025\n 
DTSTART;TZID=Europe/Berlin:20251010T114500
DTEND;TZID=Europe/Berlin:20251010T123000
LOCATION:ICP, , Seminarraum 1.079, Allmandring 3, 70569  Vaihingen 
URL;VALUE=URI:https://www.icp.uni-stuttgart.de/news/events/ESPResSo-Summer-School-Dr.-Simon-Olsson/
END:VEVENT
END:VCALENDAR