Research map / Federated learning and privacy / Differential privacy
Differentially Private Learning with Per-Sample Adaptive Clipping
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the DP-PSAC non-monotonic per-sample adaptive weight instead of constant clipping | DP training · Image classification datasets in Table 2 | test accuracy (%): 98.17\pm 0.07 -> 98.24\pm 0.07; 86.30\pm 0.21 -> 86.56\pm 0.16; 92.72\pm 0.16 -> 92.78\pm 0.13; 63.66\pm 0.05 -> 64.00\p | Helped |
| Use DP-PSAC instead of the strongest baseline | roberta-base DP training · GLUE sentence classification, epsilon=3 | test accuracy: 83.22 -> 82.74; 82.99 -> 83.36; 85.76 -> 85.83; 87.42 -> 87.48; 92.32 -> 92.43 | Mixed |
| Use DP-PSAC instead of the strongest baseline | roberta-base DP training · GLUE sentence classification, epsilon=8 | test accuracy: 83.82 -> 83.65; 83.55 -> 83.87; 86.58 -> 86.60; 87.94 -> 88.03; 92.43 -> 92.55 | Mixed |
| Replace Auto-S/NSGD's monotonically decreasing sample weight with a non-monotonic adaptive weight | DP-PSAC · Non-convex optimization (theoretical bound) | non-vanishing bound order: \mathcal{O}(r^{-1}) -> \mathcal{O}(r^{-1/2}) | Helped |
Closest papers by meaning
- Online Sensitivity Optimization in Differentially Private Learning (AAAI 2024 · similarity 0.91)
- Differentially Private SGD Without Clipping Bias: An Error-Feedback Approach (ICLR 2024 · similarity 0.90)
- Towards hyperparameter-free optimization with differential privacy (ICLR 2025 · similarity 0.89)
- Differentially private optimization for non-decomposable objective functions (ICLR 2025 · similarity 0.88)
- Delving into Differentially Private Transformer (ICML 2024 · similarity 0.88)
- Differentially Private Prototypes for Imbalanced Transfer Learning (AAAI 2025 · similarity 0.88)
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.