Title: Can machines learn all of the rules of physics, chemistry, and biology?

Date: 28th May 2021 (Friday)

Time: 14:00 pm (Central European Summer Time)

Host: prof. Joanna Trylska,

Virtual seminar: https://us02web.zoom.us/j/81220964820

Meeting ID: 812 2096 4820

To subscribe to receive announcements about CeNT UW seminars or to unsubscribe please click here

CeNT-UW-Webinars-instruction-for-attendees

Abstract:

Artificial intelligence is making many inroads in traditional areas of science. Initial efforts were complementary to traditional approaches towards interpreting and predicting data, but most recent successes are surpassing what theory and empirical methods can deliver, especially in areas where there is extensive high-quality data for machines to learn from. A striking example is the field of protein structure prediction where machine learning has led to tremendous advances solving a problem long faced with tremendous challenges. To illustrate what machine learning can deliver, recent predictions from the latest round of CASP are analyzed in depth via more conventional physics-based simulations that were initially meant to refine imperfect models by adding physical insights. The conclusion appears to be that machines learned not just all of the physics and chemistry of protein structures but even part of the biological context that modulate specific aspects. The broader implications for what may be possible in the future are profound and will be discussed based on additional examples.