Model-Agnostic FDR Control via Group Gaussian Mirror and Permutation SHAP
Jiaan Han, Junxiao Chen, Yanzhe Fu
stat.ML, cs.LG
Submitted: 2026-08-02
License: http://creativecommons.org/licenses/by-nc-sa/4.0/
The gist: Most FDR-controlled feature selection methods are designed for coordinate-wise hypotheses, where each feature has a single weight or importance score.
Terminology
Abstract
Most FDR-controlled feature selection methods are designed for coordinate-wise hypotheses, where each feature has a single weight or importance score. This abstraction fails in sequential and grouped models, where one original feature is represented by a block of sub-features, such as lags, recurrent states, or attention-based interactions. We propose a grouped-feature FDR control framework for such settings. For grouped linear models, we construct null-symmetric block-level mirror statistics with matrix-valued perturbations. For neural sequential models, we combine Permutation SHAP derivatives as model-agnostic block-level importance scores with kernel-based dependence measure. The framework is model-agnostic across network architectures, does not require specifying the covariate distribution, and reduces to Gaussian Mirror or Neural Gaussian Mirror when the block size is one. We prove FDR control for low- and high-dimensional grouped linear models and asymptotic symmetry of smoothed Permutation SHAP derivatives under fixed fitted nonlinear models. Experiments on simulated and real-world datasets show reliable FDR control and improved power under correlated grouped-feature signals.
Sources
- Near-optimal multiple testing in Bayesian linear models with finite-sample FDR control
- Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling
- Consistent Feature Selection for Analytic Deep Neural Networks
- TokenSHAP: Interpreting Large Language Models with Monte Carlo Shapley Value Estimation
- Normalizing Flows for Knockoff-free Controlled Feature Selection
- Interpretation of Prediction Models Using the Input Gradient
- DeepPINK: reproducible feature selection in deep neural networks
- Sampling Permutations for Shapley Value Estimation
- Kernel Mean Embedding of Distributions: A Review and Beyond
- Provable FDR Control for Deep Feature Selection: Deep MLPs and Beyond
- DeepDRK: Deep Dependency Regularized Knockoff for Feature Selection
- Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) Network
- CausalRivers -- Scaling up benchmarking of causal discovery for real-world time-series
- DeepLINK-T: deep learning inference for time series data using knockoffs and LSTM
Related papers
- Behavior of prediction performance metrics with rare events
- Optimal Estimation of Generic Dynamics by Path-Dependent Neural Jump ODEs
- A Posterior-Dynamics Framework for Imaging Inverse Problems with Pretrained Diffusion Priors
- Scalable Posterior Uncertainty for Flexible Density-Based Clustering
- A Transport Geometry of Benign Overfitting
- Efficient quadratic entropy with distance sketches