mlsanityData to 14 Oct 2026

Model-Agnostic FDR Control via Group Gaussian Mirror and Permutation SHAP

arXiv 2608.00989 ↗First posted 2 Aug 2026stat.ML cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

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.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 37 of its 60 references that are in the map. First posted on arXiv 2 Aug 2026.

Builds on

37 of 60 references.

References · 37

  1. 2017Attention Is All You Needcited by 60,361
  2. 2017A Unified Approach to Interpreting Model Predictionscited by 6,555
  3. 2014Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translationcited by 4,057
  4. 2014Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modelingcited by 2,367
  5. 2020Shortcut Learning in Deep Neural Networkscited by 1,969
  6. 2017Interpretation of Neural Networks is Fragilecited by 444
  7. 2018Fundamentals of Recurrent Neural Network (RNN) and Long Short-Term Memory (LSTM) Networkcited by 420
  8. 2016Kernel Mean Embedding of Distributions: A Review and Beyondcited by 364
  9. 2016Panning for Gold: Model-X Knockoffs for High-dimensional Controlled Variable Selectioncited by 121
  10. 2021Sampling Permutations for Shapley Value Estimationcited by 102
  11. 2014Controlling the false discovery rate via knockoffscited by 97
  12. 2019LassoNet: A Neural Network with Feature Sparsitycited by 82
25 more
  1. 2024TokenSHAP: Interpreting Large Language Models with Monte Carlo Shapley Value Estimationcited by 42
  2. 2018DeepPINK: reproducible feature selection in deep neural networkscited by 41
  3. 2015De-biasing the Lasso: Optimal Sample Size for Gaussian Designscited by 35
  4. 2020Consistent Feature Selection for Analytic Deep Neural Networkscited by 27
  5. 2025CausalRivers -- Scaling up benchmarking of causal discovery for real-world time-seriescited by 26
  6. 2016A knockoff filter for high-dimensional selective inferencecited by 25
  7. 2016Interpretation of Prediction Models Using the Input Gradientcited by 24
  8. 2019AFS: An Attention-based mechanism for Supervised Feature Selectioncited by 21
  9. 2018Neural Granger Causalitycited by 19
  10. 2020False Discovery Rate Control via Data Splittingcited by 12
  11. 2016The knockoff filter for FDR control in group-sparse and multitask regressioncited by 11
  12. 2019Metropolized Knockoff Samplingcited by 10
  13. 2019Controlling False Discovery Rate Using Gaussian Mirrorscited by 10
  14. 2020A Scale-free Approach for False Discovery Rate Control in Generalized Linear Modelscited by 8
  15. 2021Normalizing Flows for Knockoff-free Controlled Feature Selectioncited by 5
  16. 2021High-Dimensional Knockoffs Inference for Time Series Datacited by 5
  17. 2024FDR-Controlled Portfolio Optimization for Sparse Financial Index Trackingcited by 5
  18. 2020Neural Gaussian Mirror for Controlled Feature Selection in Neural Networkscited by 3
  19. 2020Competition-based control of the false discovery proportioncited by 3
  20. 2022Near-optimal multiple testing in Bayesian linear models with finite-sample FDR controlcited by 3
  21. 2014Generalized Twin Gaussian Processes using Sharma-Mittal Divergencecited by 2
  22. 2022Landmark Tracking in Liver US images Using Cascade Convolutional Neural Networks with Long Short-Term Memorycited by 2
  23. 2024DeepLINK-T: deep learning inference for time series data using knockoffs and LSTMcited by 2
  24. 2024DeepDRK: Deep Dependency Regularized Knockoff for Feature Selectioncited by 1
  25. 2025Provable FDR Control for Deep Feature Selection: Deep MLPs and Beyondcited by 1

Built on by

Not cited yet.