B1880
Title: Sparse-group SLOPE: Adaptive bi-level selection with FDR-control
Authors: Fabio Feser - Imperial College London (United Kingdom) [presenting]
Marina Evangelou - Imperial College London (United Kingdom)
Abstract: A new high-dimensional approach is proposed for simultaneous variable and group selection, called Sparse-group SLOPE (SGS). SGS achieves false discovery rate control at both variable and group levels by incorporating the sorted L-One penalized estimation (SLOPE) model into a sparse-group framework and exploiting grouping information. A proximal algorithm is implemented for fitting SGS that works for both Gaussian and Binomial-distributed responses. Penalty sequences specific to SGS were derived and shown to provide FDR control under orthogonal designs. Through the analysis of both synthetic and real datasets, the proposed SGS approach is found to outperform other existing lasso- and SLOPE-based models for bi-level selection and prediction accuracy. Further, the problem of model selection is investigated, with regard to FDR-control through the choice of the tuning parameter. Various model selection and noise estimation approaches for selecting the tuning parameter of the regularisation model are proposed and compared in a simulation study. Additionally, a new adaptive noise estimation procedure is proposed for SGS, termed Adaptively Scaled SGS (AS-SGS), and is an extension of the scaled lasso.