BEGIN:VCALENDAR PRODID:-//Columba Systems Ltd//NONSGML CPNG/SpringViewer/ICal Output/3.3- M3//EN VERSION:2.0 CALSCALE:GREGORIAN METHOD:PUBLISH BEGIN:VEVENT DTSTAMP:20260120T145721Z DTSTART:20260211T110000Z DTEND:20260211T120000Z SUMMARY:AI-Fun & ELLIS Invited Speaker Series | Joey Bose UID:{http://www.columbasystems.com/customers/uom/gpp/eventid/}e1dy-mkch1i av-i3c1vw DESCRIPTION:On 11 February\, we will have Joey Bose from Imperial College London.\n\nIf you cannot attend in person\, please register via the Tic ketsource link provided and you will receive the link to join the Teams session.\n\nTitle: Flow Maps and Normalizing Flows for Accelerated Gener ative Modelling of Molecules\n\nAbstract: \nThis talk will be broken int o 2 distinct parts. The first part will focus on an emerging class of ef ficient generative models known as Flow-maps that achieve state-of-the-a rt performance on few-step generation—-at a fraction of the inference co st of conventional diffusion and flow-matching based models. We will rec ap and approach the theory of Flow-maps from first principles and show h ow they can be adapted to training All-Atom Protein generative models wi th a novel Denoiser-based formulation of the Lagrangian Flow-map. This f ormulation will illuminate how existing best practices from diffusion mo dels—-such as using the Kabsch algorithm for alignment—-can be seamlessl y adopted for Flow-maps as well.\n\nThe second part of the talk will foc us on the pain points of existing generative models when applied to mode lling molecular systems. In particular\, we will discuss their applicati on to Boltzmann sampling under the framework of Boltzmann Generators\, w hich pair an exact likelihood generative model trained on biased data wi th a subsequent importance sampling step to draw statistically independe nt and consistent samples from the target Boltzmann distribution. To ac celerate the scalability of Bgs\, we will revisit classical normalizing flows in the context that offer efficient sampling and likelihoods\, but whose training via maximum likelihood is often unstable and computation ally challenging. We will introduce propose Regression Training of Norma lizing Flows (RegFlow)\, a novel and scalable regression-based training objective that bypasses the numerical instability and computational chal lenge of conventional maximum likelihood training in favour of a simple ?2-regression objective. Specifically\, RegFlow maps prior samples under our flow to targets computed using optimal transport couplings or a pre -trained continuous normalizing flow (CNF). To enhance numerical stabili ty\, RegFlow employs effective regularization strategies such as a new f orward-backward self-consistency loss that enjoys painless implementatio n. Empirically\, we demonstrate that RegFlow unlocks a broader class of architectures that were previously intractable to train for BGs with max imum likelihood. We also show that RegFlow exceeds the performance\, com putational cost\, and stability of maximum likelihood training in equili brium sampling in Cartesian coordinates of alanine dipeptide\, tripeptid e\, and tetrapeptide\, showcasing its potential in molecular systems.\n\ nBio:\nJoey Bose is an Assistant Professor of Computing at Imperial Coll ege London\, an ELLIS member\, and an Affiliate member Mila. Previously\ , he was a Post-Doctoral Fellow at University of Oxford working with Mic hael Bronstein. He completed his PhD at McGill/Mila under the supervisio n of Will Hamilton\, Gauthier Gidel\, and Prakash Panagaden. His researc h interests span Generative Modelling\, Differential Geometry for Machin e Learning with a current emphasis on geometric generative models for sc ientific applications. Previously\, he completed his Bachelors and Maste r’s degrees from the University of Toronto working on adversarial attack s against face detection and is the President and CEO of FaceShield Inc an educational platform for digital privacy for facial data. His work ha s been featured in Forbes\, New York Times\, CBC\, VentureBeat and other media outlets and was generously supported by the IVADO PhD Fellowship\ , and NSERC Post-doc Fellowship. STATUS:TENTATIVE TRANSP:TRANSPARENT CLASS:PUBLIC LOCATION:Lecture Theatre 1.3\, Kilburn Building\, Manchester END:VEVENT END:VCALENDAR