Dr. Ben Moseley
How to incorporate physical understanding into machine learning
Donnerstag 28. September 2023,
um 19:30 Uhr
ETH Hauptgebäude Hörsaal HG G5, Rämistrasse 101 und Live Stream über Zoom (Link via e-Mail)
|
|
AI and physics wisdom as seen by an AI (Bing image creator/Eschle)
|
|
Machine learning (ML) is having a profound impact on science by accelerating discoveries across physics, chemistry, and biology. Yet, entirely replacing our existing scientific workflows with purely data-driven ML models often yields unsatisfactory results. Such models struggle to generalise, require large amounts of training data, and are often seen as “black-boxes”. A more powerful approach may be to combine ML with our prior understanding of physics. Such physics-informed ML models can learn incrementally, generalise to new tasks and be more interpretable. In this talk we give an overview of the different ways physical principles can be combined with machine learning, and the impact this is having on scientific research.
[Wegbeschreibungen]
[Veranstaltungs-Uebersicht]
|