Sufficient Dimension Reduction and Modeling Responses Conditioned on Covariates: An Integrated Approach via Convex Optimization

Given observations of a collection of covariates and responses , sufficient dimension reduction (SDR) techniques aim to identify a mapping with such that is independent of . The image summarizes the relevant information in a potentially large number of covariates that influence the responses . In many contemporary settings, the number of responses is also quite large, in addition to a large number of covariates. This leads to the challenge of fitting a succinctly parameterized statistical model to , which is a problem that is usually not addressed in a traditional SDR framework. In this paper, we present a computationally tractable convex relaxation based estimator for simultaneously (a) identifying a linear dimension reduction of the covariates that is sufficient with respect to the responses, and (b) fitting several types of structured low-dimensional models -- factor models, graphical models, latent-variable graphical models -- to the conditional distribution of . We analyze the consistency properties of our estimator in a high-dimensional scaling regime. We also illustrate the performance of our approach on a newsgroup dataset and on a dataset consisting of financial asset prices.
View on arXiv