Skip to content. | Skip to navigation

Personal tools

Theme for TIFR Centre For Applicable Mathematics, Bangalore


You are here: Home / Events / Colloquium : A new score-based transport method for sampling

Colloquium : A new score-based transport method for sampling

Nisha Chandramoorthy (Assistant Professor, Georgia Tech)
Nisha Chandramoorthy (Assistant Professor, Georgia Tech)
When Jun 27, 2023
from 03:00 PM to 04:00 PM
Where LH-111 (TIFR CAM)
Add event to calendar vCal


We discuss a new construction of a solution to the measure transport problem. This solution is defined as a zero of an infinite-dimensional score-matching problem. We develop an infinite-dimensional generalization of a Newton method to find this zero, which also serves as its constructive existence proof. We define a score operator that gives the difference of the score -- gradient of logarithm of density -- of a transported distribution from the target score. The Newton method is iterative, enjoys fast convergence under smoothness assumptions, and does not make a parametric ansatz on the transport map. It is appropriate for the variational inference setting, where the score is known, and for sampling certain chaotic dynamical systems, where a conditional score can be calculated even in the absence of a statistical model for the target. Fast computation of scores in this setting is discussed along with the roadmap to applying the transport algorithm to Bayesian filtering. Joint work with Youssef Marzouk.

Filed under: