BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Steering Generative Models for Discovery - Andreas Krause (ETH Zü
 rich)
DTSTART:20260320T101500Z
DTEND:20260320T111500Z
UID:TALK244486@talks.cam.ac.uk
DESCRIPTION:Large-scale generative models\, which can be steered to optimi
 ze specific objectives\, have yielded remarkable successes in scientific a
 pplications such as de-novo protein design.&nbsp\; However\, a central cha
 llenge in scientific discovery is to explore beyond the domain well-repres
 ented by the training data. &nbsp\;\nIn this talk\, I will present recent 
 work leveraging ideas from reinforcement learning and stochastic optimal c
 ontrol to steer generative models for novelty-seeking generative discovery
 .&nbsp\; In particular\, I will introduce Flow Density Control\, a flexibl
 e framework for steering flow- and diffusion-based generative models that 
 captures diverse use-cases\, including maximum entropy manifold exploratio
 n and tail-aware generative optimization.&nbsp\; I will also discuss how v
 erifiers (assessing\, e.g.\, physical plausibility) can be utilized to con
 strain and guide the exploration process.&nbsp\; I will motivate and illus
 trate the approaches on several examples from molecular design.&nbsp\;
LOCATION:Seminar Room 1\, Newton Institute
END:VEVENT
END:VCALENDAR