Research map / Federated learning and privacy / Differential privacy
Private Adaptive Optimization with Side information
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 |
|---|---|---|---|
| Precondition gradients with public-data statistics before adding privacy noise | Private optimization · IMDB | accuracy: 0.62 -> 0.80 | Helped |
| Precondition gradients with public-data statistics before adding privacy noise | Private optimization · StackOverflow | accuracy: 0.33 -> 0.40 | Helped |
| Precondition gradients with public-data statistics before adding privacy noise | Private optimization · MNIST | loss: 0.039 -> 0.036 | Helped |
| Precondition gradients using side information before privatization | Deep autoencoder with private training · MNIST | loss ×100 (σ=1): 3.697 { (.020)} -> 3.566 { (.008)} | Helped |
| Precondition gradients using side information before privatization | Deep autoencoder with private training · MNIST | loss ×100 (σ=0.75): 3.286 { (.016)} -> 3.158 { (.003)} | Helped |
| Add inverse TF-IDF side information for preconditioning | AdaDPS with private logistic regression · IMDB BoW | accuracy: 0.68 { (.01)} -> 0.75 { (.01)} | Helped |
| Add inverse TF-IDF side information for preconditioning | AdaDPS with private logistic regression · IMDB TF-IDF | accuracy: 0.65 { (.01)} -> 0.80 { (.00)} | Helped |
| Use OOD public data instead of ID public data (ablation) | AdaDPS · IMDB | accuracy: 0.80 -> 0.79 | No clear effect |
| Use OOD public data instead of ID public data (ablation) | AdaDPS · StackOverflow | accuracy: 0.40 -> 0.40 | No clear effect |
| Replace public data with heuristic side information (ablation) | AdaDPS · IMDB | accuracy: 0.80 -> 0.75 | Hurt |
| Replace public data with heuristic side information (ablation) | AdaDPS · StackOverflow | accuracy: 0.40 -> 0.41 | Helped |
| Use related OOD public data for preconditioning (ablation) | Private logistic regression · IMDB | accuracy: 0.63 -> 0.79 | Helped |
Closest papers by meaning
- Certification for Differentially Private Prediction in Gradient-Based Training (ICML 2025 · similarity 0.89)
- Online Sensitivity Optimization in Differentially Private Learning (AAAI 2024 · similarity 0.88)
- Differentially Private Stochastic Gradient Descent with Fixed-Size Minibatches: Tighter RDP Guarantees with or without Replacement (NeurIPS 2024 · similarity 0.88)
- DiSK: Differentially Private Optimizer with Simplified Kalman Filter for Noise Reduction (ICLR 2025 · similarity 0.88)
- DOPPLER: Differentially Private Optimizers with Low-pass Filter for Privacy Noise Reduction (NeurIPS 2024 · similarity 0.88)
- Differentially Private Domain Adaptation with Theoretical Guarantees (ICML 2024 · 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.