Research map / Federated learning and privacy / Differential privacy
DPNAS: Neural Architecture Search for Deep Learning with Differential Privacy
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 |
|---|---|---|---|
| search architectures with DPNAS instead of using CNN-Tanh | DPNASNet · MNIST, ε=3, δ=1×10⁻⁵ | accuracy: 98.10 -> 98.57 | Helped |
| search architectures with DPNAS instead of using CNN-Tanh | DPNASNet · FashionMNIST, ε=3, δ=1×10⁻⁵ | accuracy: 86.25 -> 88.09 | Helped |
| search architectures with DPNAS instead of using CNN-Tanh | DPNASNet · CIFAR-10, ε=3, δ=1×10⁻⁵ | accuracy: 59.07 -> 68.33 | Helped |
| use DPNAS search space instead of DARTS or NASNet search spaces (ablation) | DPNAS search process · CIFAR-10, ε=3, δ=1×10⁻⁵ | accuracy | Helped |
| train sampled architectures with DP-aware training instead of ordinary SGD (ablation) | DPNAS search process · CIFAR-10, ε=3, δ=1×10⁻⁵ | accuracy | Helped |
| use RL-based search instead of random search (ablation) | DPNAS search process · CIFAR-10 search | accuracy | Helped |
| replace all searched activations with ReLU (ablation) | DPNASNet · CIFAR-10, ε=3, δ=1×10⁻⁵ | accuracy: 68.33 -> 66.36 | Hurt |
| replace all searched activations with SELU (ablation) | DPNASNet · CIFAR-10, ε=3, δ=1×10⁻⁵ | accuracy: 68.33 -> 68.06 | Hurt |
| replace all searched activations with Tanh (ablation) | DPNASNet · CIFAR-10, ε=3, δ=1×10⁻⁵ | accuracy: 68.33 -> 65.31 | Hurt |
| replace all searched activations with linear activation (ablation) | DPNASNet · CIFAR-10, ε=3, δ=1×10⁻⁵ | accuracy: 68.33 -> 66.81 | Hurt |
| replace all searched activations with Hardtanh (ablation) | DPNASNet · CIFAR-10, ε=3, δ=1×10⁻⁵ | accuracy: 68.33 -> 64.80 | Hurt |
| replace all searched activations with Sigmoid (ablation) | DPNASNet · CIFAR-10, ε=3, δ=1×10⁻⁵ | accuracy: 68.33 -> 58.09 | Hurt |
Closest papers by meaning
- Towards hyperparameter-free optimization with differential privacy (ICLR 2025 · similarity 0.88)
- Delving into Differentially Private Transformer (ICML 2024 · similarity 0.87)
- Noise-Aware Differentially Private Regression via Meta-Learning (NeurIPS 2024 · similarity 0.87)
- Differentially Private Bias-Term Fine-tuning of Foundation Models (ICML 2024 · similarity 0.86)
- Revisiting Differentially Private ReLU Regression (NeurIPS 2024 · similarity 0.85)
- Convex Approximation of Two-Layer ReLU Networks for Hidden State Differential Privacy (NeurIPS 2025 · similarity 0.85)
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.