Research map / Optimization / Implicit bias and generalization
Feature Grouping as a Stochastic Regularizer for High-Dimensional Structured Data
Sergul Aydore, Bertrand Thirion, Gael Varoquaux
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use feature grouping instead of the strongest conventional MLP regularizer | MLP on Olivetti Faces · No noise | test accuracy (%): 89.55 \pm 0.88 -> 85.45 \pm 1.08 | Hurt |
| Use feature grouping instead of the strongest conventional MLP regularizer | MLP on HCP-small · No noise | test accuracy (%): 88.31 {\pm 0.14} -> 87.36 \pm 0.57 | Hurt |
| Use feature grouping instead of the strongest conventional MLP regularizer | MLP on Olivetti Faces · Medium and high noise | test accuracy (%): 61.82 \pm 1.14 -> 80.91 {\pm 1.02}; 34.02 \pm 1.48 -> 68.79 {\pm 1.04} | Improved |
| Use feature grouping instead of the strongest conventional MLP regularizer | MLP on HCP-small · Medium and high noise | test accuracy (%): 80.66 \pm 0.22 -> 83.75 {\pm 0.35}; 69.81 \pm 0.56 -> 76.45 {\pm 0.52} | Improved |
| Vary samples used per projection and number of projection matrices (ablation) | Feature grouping with logistic regression · HCP-small, 330 samples | accuracy: 79.88 \pm 0.53; 78.28 \pm 0.54; 79.24 \pm 0.66; 78.05 \pm 1.04; 78.82 \pm 0.79; 78.84 \pm 0.82; 77.97 \pm 0.823; 78.50 \ | No clear effect |
| Use feature grouping rather than conventional regularizers as the training sample count decreases | Logistic regression and MLP on HCP · Small-sample settings | test accuracy | Improved |
Closest papers by meaning
- Fast structure learning with modular regularization (NeurIPS 2019 · similarity 0.86)
- Unsupervised Feature Selection Through Group Discovery (AAAI 2026 · similarity 0.85)
- Network-to-Network Regularization: Enforcing Occam's Razor to Improve Generalization (NeurIPS 2021 · similarity 0.85)
- Structured Low-Rank Matrix Factorization: Global Optimality, Algorithms, and Applications (TPAMI 2019 · similarity 0.85)
- Generalization Performance of Ensemble Clustering: From Theory to Algorithm (ICML 2025 · similarity 0.84)
- Implicit Regularization Paths of Weighted Neural Representations (NeurIPS 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
