Online Program Home
My Program

Abstract Details

Activity Number: 298 - Model/Variable Selection and Model Evaluation
Type: Contributed
Date/Time: Tuesday, July 30, 2019 : 8:30 AM to 10:20 AM
Sponsor: Biometrics Section
Abstract #305206
Title: Integrative Multi-View Regression: Statistical Inference with De-Biased and Scaled Composite Nuclear Norm Penalization
Author(s): Xiaokang Liu* and Kun Chen
Companies: University of Connecticut and University of Connecticut
Keywords: bias correction; chi-squared distribution; group inference; multi-view learning; nuclear norm penalization; integrative multivariate analysis

Integrative reduced-rank regression (iRRR) studies the predictive association between multivariate response and multi-view predictor sets. It nicely bridges sparse, group-sparse and low-rank models by exploiting a composite nuclear norm penalization. In this work, we derive an approximate chi-squared test on groups of predictors (coefficient sub-matrices), based on the estimators of both the noise level and the coefficient matrix after de-biasing from a scaled version of iRRR. The proposed method enables valid inference on view selection in multi-view learning, and it also subsumes inference procedures for lasso, group lasso and nuclear norm penalized regression estimators as special cases. Furthermore, the method is extended to allow for flexible linear constraints on the coefficients, which enables inference for a multivariate log-contrast model with multi-view compositional predictors. Simulation studies confirm the validity of the proposed inference procedure. We analyze data collected from a preterm infant study, to identify taxa of gut microbiome compositions during early postnatal period that may impact later neurobehavioral outcomes of preterm infants.

Authors who are presenting talks have a * after their name.

Back to the full JSM 2019 program