Research map / Probabilistic and scientific modeling / Molecules and proteins
Data Minimization at Inference Time
Cuong Tran, Ferdinando Fioretto
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 personalized MinDRel feature disclosure instead of requiring all sensitive features | Inference on the evaluated learning tasks · Various benchmarks | prediction accuracy: 10\% of their information while maintaining the same accuracy level | No clear effect |
| Use F-Score selection rather than Importance or Random selection (ablation) | MinDRel · Linear-classifier evaluations | data leakage | Improved |
| Increase the allowed failure probability δ (ablation) | MinDRel · Linear-classifier evaluations | relative data leakage | Improved |
| Increase the allowed failure probability to δ ≤ 0.1 (ablation) | MinDRel · Various datasets, nonlinear classifiers | sensitive features disclosed: 5\% | Improved |
| Use MinDRel with δ = 0 rather than the all-features baseline | Nonlinear classifiers · Various datasets | accuracy: 0.005\% | Hurt |
| Use the Gaussian assumption rather than Bayesian NN to model conditional feature distributions (ablation) | MinDRel · Bank, Income, Credit, Insurance (|S| = 5) | training time (minutes): 204, 375, 125, 90 -> 0.01, 0.02, 0.02, 0.01 | Improved |
| Use the Gaussian assumption rather than Bayesian NN to model conditional feature distributions (ablation) | MinDRel · Bank, Income, Credit, Insurance (|S| = 5, δ = 0) | inference time on test set (minutes): 40, 254, 220, 34 -> 15, 78, 66, 9 | Improved |
| Use the Gaussian assumption rather than Bayesian NN to model conditional feature distributions (ablation) | MinDRel · Credit dataset (|S| = 5) | accuracy and data leakage | No clear effect |
| Reveal features in descending importance order rather than using F-Score (ablation) | MinDRel · Linear-classifier evaluations | data leakage | Hurt |
| Reveal features in random order rather than using F-Score (ablation) | MinDRel · Linear-classifier evaluations | data leakage | Hurt |
Closest papers by meaning
- The Fundamental Limits of Least-Privilege Learning (ICML 2024 · similarity 0.83)
- It's My Data Too: Private ML for Datasets with Multi-User Training Examples (ICML 2025 · similarity 0.82)
- Beyond Memorization: Violating Privacy via Inference with Large Language Models (ICLR 2024 · similarity 0.82)
- I Prefer Not to Say: Protecting User Consent in Models with Optional Personal Data (AAAI 2024 · similarity 0.82)
- Learning from Aggregate responses: Instance Level versus Bag Level Loss Functions (ICLR 2024 · similarity 0.82)
- Prediction without Preclusion: Recourse Verification with Reachable Sets (ICLR 2024 · similarity 0.81)
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.
