Knockoffs-based False Discovery Rate Control and Simplification for Deep Neural Networks

2026-06-02

AI Summary

This study addresses the prevalence of redundant parameters and irrelevant input variables in deep neural networks, which often lead to computational inefficiency, reduced model interpretability, and poor statistical reliability. To tackle this issue, the authors systematically integrate the knockoffs framework into deep learning for the first time, combining regularized network architectures with false discovery rate (FDR) control theory. They propose three hierarchical variable screening strategies: single-layer filtering, multi-layer filtering, and variable weight aggregation filtering. Evaluated on multiple benchmark datasets, the proposed methods significantly reduce redundant features while strictly controlling FDR, yielding simplified models that outperform existing approaches in terms of predictive performance, computational efficiency, interpretability, and statistical guarantees.

Abstract

The deep neural network is a widely used framework in machine learning that has been widely applied in various fields. However, deep neural networks often involve a large number of parameters and inputs, many of which may be irrelevant to the goal or true output. These parameters and \textcolor{black}{input variables} not only increase computational complexity, but also contribute to additional computational cost. One solution to this problem is knockoff methods, which have proven successful in controlling false discovery rates in high-dimensional regression. Building on the knockoff methods and using the regularised neural network, this paper proposes three variable screening methods under the condition of controlling false discovery rates: \textit{one layer filter}, \textit{multiple layers filter}, \textit{variable weight aggregation filter}. In comparison with existing algorithms, we find that our algorithms show satisfactory performance.

Problem

• deep neural networks
• false discovery rate
• variable selection
• knockoffs
• high-dimensional data

Innovation

• knockoffs
• false discovery rate control
• variable screening
• regularized neural networks
• deep learning interpretability