革新知能統合研究センター

*This is a hybrid seminar and AIP researchers can join onsite at Meeting Room 4.

Speaker: Prof. Michael Choi (National University of Singapore and Yale-NUS College)

Title: From reversiblizations of non-reversible Markov chains to landscape modification

Abstract: In the first part of the talk, we provide a brief literature survey on various reversiblization techniques for analyzing non-reversible Markov chains. This include the additive, multiplicative and Metropolis-Hastings reversiblizations, their spectral and geometric relationships, and how these reversiblizations offer insights to the analysis of the non-reversible Markov chains.
In the second part of the talk, inspired by the reversiblization techniques in the first part, we develop a new acceleration method that we call landscape modification, which can possibly be applied to accelerate gradient-based sampling or optimization algorithms.