Biostatistics Seminar speaker, Ali Shojaie, PhD, Associate Professor, Dept. of Biostatistics, University of Washington, will present, “Flexibility in High Dimensions: Sparse Additive Models with Adaptive Truncation via a Convex Hierarchical Penalty”.

 

ABSTRACT: We consider the problem of nonparametric regression with a potentially large number of covariates. We propose a convex penalized estimation framework that is particularly well-suited for high-dimensional sparse additive models. Existing sparse additive modeling approaches assume that all additive components have the same level of complexity and are thus not data-adaptive. In contrast, the proposed approach selects the appropriate level of complexity for each additive component data-adaptively . Importantly, this flexibility is achieved without sacrificing computational efficiency: We demonstrate that the proposed approach scales similarly to the LASSO with the number of covariates and samples size. We demonstrate these properties through empirical studies on both real and simulated datasets and show that our estimator converges at the minimax rate.