DNAS Seminar Series

From “Negative Probabilities” To Machine Learning

Date & Time

Date: Oct 21 Tuesday

Time: 4:15-5:15pm Keynote Lecture: From “Negative Probabilities” To Machine Learning

          5:15-6:15pm DUKE Master program introduction in Materials Science & Engineering

Venue: IB 1046& Zoom

Zoom ID:  789 843 9496

Please register in advance to reserve a seat, thank you!

Speaker

Prof. Patrick Charbonneau

Professor of Physics, Duke University

Lead Editor of Physical Review E

Abstract

The random clusters introduced by Fortuin and Kasteleyn (FK) were leveraged first by Swendsen and Wang and then by Wolff to formulate remarkably efficient sampling schemes that weaken the critical slowing down of Monte Carlo simulations. By contrast, in models with frustration no standard way to produce comparable sampling gain has yet been identified. In this talk, I present our recent study of various such models on Bethe lattices, which has allowed us to identify the proper generalization of FK clusters. However, a standard, constructive cluster scheme is then inoperable. In addition, the frustration range over which these generalized clusters are even definable. The potential of leveraging machine learning to enhance sampling in these systems will also be discussed.

BIO

Patrick Charbonneau is Professor of Physics at Duke University and Lead Editor of Physical Review E. Trained at McGill (BSc) and Harvard (PhD), with postdoctoral work at AMOLF, in Amsterdam, he specializes in soft matter and statistical physics, focusing on glass formation and self-assembly in frustrated systems. He also advances the history of science through projects on replica symmetry breaking, quantum chemistry, and confectionery, bridging cultural history with material science