Research map / Learning theory and algorithms / Sample complexity
Characterization of Gaussian Universality Breakdown in High-Dimensional Empirical Risk Minimization
Mohamed Chiheb Yaakoubi, Cosme Louart, Malik Tiomoko, Zhenyu Liao
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 elastic loss (1-eta)*squared + eta*absolute with eta>0 instead of pure squared loss (ablation) | Gaussian surrogate performance prediction · Teacher-student regression, bimodal features, p=30, n=50 | test error prediction | Hurt |
| pure squared loss (eta=0) with non-Gaussian bimodal features (ablation) | Gaussian surrogate performance prediction · Teacher-student regression, bimodal features | test error prediction | No clear effect |
| replace data score by Gaussian score assumption (ablation) | Classification error prediction · MNIST classification, varying regularization | classification error | Hurt |
| use non-Gaussian CGMT theoretical prediction instead of Gaussian-data prediction | ERM generalization error prediction · Elastic-loss regression and MNIST | test/classification error | Improved |
Closest papers by meaning
- The Breakdown of Gaussian Universality in Classification of High-dimensional Linear Factor Mixtures (ICLR 2025 · similarity 0.89)
- On the Asymptotic Distribution of the Minimum Empirical Risk (ICML 2024 · similarity 0.89)
- Information Theoretic Guarantees for Empirical Risk Minimization with Applications to Model Selection and Large-Scale Optimization (ICML 2018 · similarity 0.88)
- Universality laws for Gaussian mixtures in generalized linear models (NeurIPS 2023 · similarity 0.88)
- Differentially Private Empirical Risk Minimization with Smooth Non-Convex Loss Functions: A Non-Stationary View (AAAI 2019 · similarity 0.88)
- On the Efficiency of ERM in Feature Learning (NeurIPS 2024 · similarity 0.87)
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.
