Schrödinger bridge and Trajectory Inference (Part 1)

Thumbnail

Event details

Date 16.03.2023
Hour 14:0015:00
Speaker Chizat Lénaic (EPFL)  
Location
Bernoulli Center
Category Conferences - Seminars
Event Language English

Abstract:
We consider statistical and computation methods to infer trajectories of a stochastic process from snapshots of its temporal marginals. This problem arises for instance in the analysis of single cell RNA-sequencing data. The goal of these lectures is to present and understand the estimator proposed by [Lavenant et al. 2020] which searches for the diffusion process that fits the observations with minimal entropy relative to a Wiener process. This estimator comes with consistency guarantees—for a suitable class of ground truth processes—and lends itself to computational methods with global optimality guarantees. Its analysis is the occasion to review important tools from entropic optimal transport — aka the Schrödinger bridge problem.
 

Practical information

  • General public
  • Free

Contact

  • Prof. Xue-Mei Li

Tags

Working Seminar

Event broadcasted in

Share