Modeling and Simulation Group Meeting Old
Uncovering hidden signals and causal effects from data through optimal transport
Speaker: Esteban Tabak, Courant
Date: Thursday, March 4, 2021, 12:30 p.m.
Two related general problems in science-oriented data analysis are to extract from the data information on unobserved factors that may explain the behavior of the quantities of interest, and to infer and quantify causal relations among variables. One example of the former is uncovering from climatological data hidden patterns, such as El Niño Southern Oscillation, which may explain a large share of the observed variability. One example of the latter is to infer from observational data the effect of a medical treatment in the presence of confounders, such as the severity of the health condition, which may affect both the treatment adopted and its effect.
In recent years, optimal transport and related problems, such as the Wasserstein barycenter problem, have been shown to provide a very natural framework for the analysis of data, particularly regarding the estimation and simulation of conditional probabilities. This talk/conversation will describe this framework and discuss how it applies to uncovering hidden signals and causal effects.