Steering Generative Models for Discovery

Add to your list(s) Download to your calendar using vCal

• Andreas Krause (ETH Zürich)
• Friday 20 March 2026, 10:15-11:15
• Seminar Room 1, Newton Institute.

If you have a question about this talk, please contact nobody.

SCLW02 - Reinforcement Learning for Science: Discovery and Automation

Large-scale generative models, which can be steered to optimize specific objectives, have yielded remarkable successes in scientific applications such as de-novo protein design.  However, a central challenge in scientific discovery is to explore beyond the domain well-represented by the training data.   In this talk, I will present recent work leveraging ideas from reinforcement learning and stochastic optimal control to steer generative models for novelty-seeking generative discovery.  In particular, I will introduce Flow Density Control, a flexible framework for steering flow- and diffusion-based generative models that captures diverse use-cases, including maximum entropy manifold exploration and tail-aware generative optimization.  I will also discuss how verifiers (assessing, e.g., physical plausibility) can be utilized to constrain and guide the exploration process.  I will motivate and illustrate the approaches on several examples from molecular design. 

This talk is part of the Isaac Newton Institute Seminar Series series.

This talk is included in these lists:

• All CMS events
• Featured lists
• INI info aggregator
• Isaac Newton Institute Seminar Series
• School of Physical Sciences
• Seminar Room 1, Newton Institute
• bld31
• dh539

Note that ex-directory lists are not shown.