Research map / Federated learning and privacy / Differential privacy
Efficient Public Verification of Private ML via Regularization
Zoë Ruha Bell, Anvith Thudi, Olive Franzese, Nicolas Papernot, Shafi Goldwasser
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 |
|---|---|---|---|
| Run verification over a LAN instead of locally (ablation) | Phased ERM certification · MNIST full verification | time (hr): 5.54 -> 5.73 | Hurt |
| Run verification over a WAN (200 Mb) instead of locally (ablation) | Phased ERM certification · MNIST full verification | time (hr): 5.54 -> 6.70 | Hurt |
| Run verification over a WAN (100 Mb) instead of locally (ablation) | Phased ERM certification · MNIST full verification | time (hr): 5.54 -> 6.88 | Hurt |
| Replace prior DP-SGD certification with modified phased ERM certification | MNIST logistic regression DP verification · MNIST | verification time (hr): 100 -> 3 | Improved |
| Replace DP-SGD gradient checks with phased ERM gradient checks | MNIST DP certification · MNIST | gradients verified: 255,280 -> 60,000 | Improved |
| Vary epsilon from 1.665 to 0.166 (ablation) | Phased ERM certification · MNIST full verification | time (hr): 3.10 -> 3.19 | No clear effect |
| Increase noise sigma from 0.1 to 10 (ablation) | Gaussian commitment · Noise generation | time (ms/sample): 0.071 -> 0.94 | Hurt |
| Increase features from 512 to 2048 (ablation) | Gradient verification · Gradient verification | time (ms/gradient): 109 -> 432 | Hurt |
| Require small gradient norm and divide delta across phases in phased ERM | Phased ERM · MNIST logistic regression training | training time (s): 14.37 -> 19.24 | Mixed |
Closest papers by meaning
- Optimal Differentially Private Model Training with Public Data (ICML 2024 · similarity 0.92)
- Faster Algorithms for User-Level Private Stochastic Convex Optimization (NeurIPS 2024 · similarity 0.89)
- Provable Privacy with Non-Private Pre-Processing (ICML 2024 · similarity 0.89)
- Public-data Assisted Private Stochastic Optimization: Power and Limitations (NeurIPS 2024 · similarity 0.89)
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.89)
- Differentially Private Stochastic Gradient Descent with Fixed-Size Minibatches: Tighter RDP Guarantees with or without Replacement (NeurIPS 2024 · similarity 0.89)
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.
