A0889
Title: High-dimensional variable selection in non-linear mixed-effects models
Authors: Guillaume Kon Kam King - Université Paris-Saclay, INRAE (France) [presenting]
Abstract: High-dimensional data (many more covariates than observations) are now commonly analyzed. However, there are few tools for high-dimensional variable selection when the data are observations collected repeatedly on several individuals, and even fewer when the model is nonlinear. Thus, a high-dimensional covariate selection method is developed for nonlinear mixed-effects models, which are natural tools for analyzing data of this nature. More precisely, it is proposed to use a spike and slab prior for variable selection, coupled with a stochastic approximation version of the EM algorithm for scalability. Targeting the maximum a posteriori, the proposed approach is much faster than classical MCMC procedures and shows very good selection performance on simulated data.
