Research map / Federated learning and privacy / Privacy attacks and secure learning
Generalization in Generative Adversarial Networks: A Novel Perspective from Privacy Protection
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 |
|---|---|---|---|
| Add spectral normalization | DCGAN with JS objective · LFW | Gap: 0.581 -> 0.106 | Helped |
| Add spectral normalization | DCGAN with JS objective · LFW | AUC: 0.729 -> 0.506 | Helped |
| Add weight clipping | DCGAN with JS objective · LFW | Gap: 0.581 -> 0.113 | Helped |
| Add spectral normalization | DCGAN with JS objective · IDC | AUC: 0.531 -> 0.508 | Helped |
| Add spectral normalization | DCGAN with JS objective · IDC | F1: 0.445 -> 0.416 | Helped |
| Use Wasserstein objective without weight clipping | DCGAN · LFW and IDC | training stability: N/A | Hurt |
| Use gradient penalty with JS objective | DCGAN · LFW and IDC | training stability: N/A | Hurt |
| Use original weight normalization | DCGAN · LFW and IDC | training stability | Hurt |
| Add orthonormal regularization | DCGAN · IDC | membership attack F1: 0.402 | Helped |
| Use spectral normalization instead of weight clipping | DCGAN with JS objective · LFW | Inception score: 3.112 vs 3.104 | Mixed |
| Use black-box rather than white-box attack | Original DCGAN · LFW | AUC: 0.729 -> 0.549 | Hurt |
| Add spectral normalization under black-box attack | DCGAN · LFW | AUC: 0.549 -> 0.497 | Helped |
Closest papers by meaning
- PrAda-GAN: A Private Adaptive Generative Adversarial Network with Bayes Network Structure (AAAI 2026 · similarity 0.90)
- PATE-GAN: Generating Synthetic Data with Differential Privacy Guarantees (ICLR 2019 · similarity 0.90)
- Improving Generalization and Stability of Generative Adversarial Networks (ICLR 2019 · similarity 0.90)
- Auditing Privacy Defenses in Federated Learning via Generative Gradient Leakage (CVPR 2022 · similarity 0.89)
- GS-WGAN: A Gradient-Sanitized Approach for Learning Differentially Private Generators (NeurIPS 2020 · similarity 0.88)
- Privacy without Noisy Gradients: Slicing Mechanism for Generative Model Training (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.