Research map / Learning theory and algorithms / Kernels and regression theory
Parameters or Privacy: A Provable Tradeoff Between Overparameterization and Membership Inference
Jasper Tan, Blake Mason, Hamid Javadi, Richard Baraniuk
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 |
|---|---|---|---|
| increase number of parameters p (gamma above 2) | minimum-norm least squares · Gaussian linear regression | membership advantage | Hurt |
| make model only slightly overparameterized (gamma at or below 2) | minimum-norm least squares · Gaussian linear regression | membership advantage | Improved |
| add ridge regularization in highly overparameterized regime | ridge regression · Gaussian linear regression | membership advantage | Hurt |
| reduce number of features (feature reduction) | linear regression · Gaussian linear regression | membership advantage | Improved |
| add Gaussian noise to outputs of non-member queries | linear regression with all features · Gaussian linear regression | membership advantage | Improved |
| compare feature reduction to noise addition on privacy-utility trade-off | linear regression · Gaussian linear regression | membership advantage | No clear effect |
| increase number of frequency components p | interpolating regressor · noise-free time-series regression | membership advantage | Hurt |
| increase number of parameters in highly overparameterized regime | random ReLU features regression · random ReLU features | membership advantage | Hurt |
Closest papers by meaning
- When Does Data Augmentation Help With Membership Inference Attacks? (ICML 2021 · similarity 0.88)
- Fundamental Limits of Membership Inference Attacks on Machine Learning Models (JMLR 2025 · similarity 0.88)
- Assessing Per-Sample Membership Inference Vulnerability without Retraining (NeurIPS 2026 · similarity 0.86)
- Disparate Impact on Group Accuracy of Linearization for Private Inference (ICML 2024 · similarity 0.85)
- Mitigating Privacy Risk in Membership Inference by Convex-Concave Loss (ICML 2024 · similarity 0.85)
- Impact of Dataset Properties on Membership Inference Vulnerability of Deep Transfer Learning (NeurIPS 2025 · similarity 0.85)
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.
