Neural Gaussian Mirror for Controlled Feature Selection in Neural Networks
This addresses the need for interpretable feature selection in scientific applications where DNNs lack transparency, offering a controlled error rate method.
The paper tackles the problem of feature importance identification in deep neural networks by introducing neural Gaussian mirrors (NGMs), which use mirrored features and mirror statistics to evaluate importance, resulting in controlled error rates at a predefined level and high selection power even with correlated features.
Deep neural networks (DNNs) have become increasingly popular and achieved outstanding performance in predictive tasks. However, the DNN framework itself cannot inform the user which features are more or less relevant for making the prediction, which limits its applicability in many scientific fields. We introduce neural Gaussian mirrors (NGMs), in which mirrored features are created, via a structured perturbation based on a kernel-based conditional dependence measure, to help evaluate feature importance. We design two modifications of the DNN architecture for incorporating mirrored features and providing mirror statistics to measure feature importance. As shown in simulated and real data examples, the proposed method controls the feature selection error rate at a predefined level and maintains a high selection power even with the presence of highly correlated features.