Research map / Federated learning and privacy / Differential privacy
Enabling Fast Differentially Private SGD via Just-in-Time Compilation and Vectorization
Pranav Shankar Subramani, Nicholas Vadivelu, Gautam Kamath
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 JAX or Custom TFP for private training | DPSGD · MNIST CNN | median runtime per epoch: 6.50 -> 0.53 | Improved |
| Use JAX or Custom TFP for private training | DPSGD · IMDb LSTM | median runtime per epoch: 407 -> 8.2 | Improved |
| Add VMAP (ablation) | JAX DPSGD · Adult logistic regression | median runtime per epoch: 1040 -> 16.4 | Improved |
| Add JIT (ablation) | JAX DPSGD · Adult logistic regression | median runtime per epoch: 1040 -> 1.08 | Improved |
| Use both VMAP and JIT (ablation) | JAX DPSGD · Adult logistic regression | median runtime per epoch: 1040 -> 0.231 | Improved |
| Add VMAP (ablation) | JAX DPSGD · CIFAR10 CNN | median runtime per epoch: 3840 -> 64.8 | Improved |
| Add JIT (ablation) | JAX DPSGD · CIFAR10 CNN | median runtime per epoch: 3840 -> 84.0 | Improved |
| Use both VMAP and JIT (ablation) | JAX DPSGD · CIFAR10 CNN | median runtime per epoch: 3840 -> 7.28 | Improved |
| Add XLA compilation (ablation) | Custom TFP DPSGD · Adult FCNN | median runtime per epoch: 12.7 -> 1.83 | Improved |
| Use graph-mode vectorization (ablation) | Custom TFP DPSGD · Adult FCNN | median runtime per epoch: 12.7 -> 0.770 | Improved |
| Add vectorization to XLA (ablation) | Custom TFP DPSGD · Adult FCNN | median runtime per epoch: 1.83 -> 0.209 | Improved |
| Add XLA compilation (ablation) | Custom TFP DPSGD · MNIST CNN | median runtime per epoch: 32.5 -> 19.5 | Improved |
Closest papers by meaning
- The importance of feature preprocessing for differentially private linear optimization (ICLR 2024 · similarity 0.88)
- Differentially Private SGD Without Clipping Bias: An Error-Feedback Approach (ICLR 2024 · similarity 0.88)
- Online Sensitivity Optimization in Differentially Private Learning (AAAI 2024 · similarity 0.87)
- Private Zeroth-Order Optimization with Public Data (NeurIPS 2025 · similarity 0.87)
- Differentially Private Stochastic Gradient Descent with Fixed-Size Minibatches: Tighter RDP Guarantees with or without Replacement (NeurIPS 2024 · similarity 0.87)
- Nearly Tight Black-Box Auditing of Differentially Private Machine Learning (NeurIPS 2024 · similarity 0.86)
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.
