Abstract
Feature selection is central to contemporary high-dimensional data analysis.Grouping structure among features arises naturally in various scientificproblems. Many methods have been proposed to incorporate the grouping structureinformation into feature selection. However, these methods are normallyrestricted to a linear regression setting. To relax the linear constraint, wecombine the deep neural networks (DNNs) with the recent Knockoffs technique,which has been successful in an individual feature selection context. Wepropose Deep-gKnock (Deep group-feature selection using Knockoffs) as amethodology for model interpretation and dimension reduction. Deep-gKnockperforms model-free group-feature selection by controlling group-wise FalseDiscovery Rate (gFDR). Our method improves the interpretability andreproducibility of DNNs. Experimental results on both synthetic and real datademonstrate that our method achieves superior power and accurate gFDR controlcompared with state-of-the-art methods.