BriefGPT.xyz
Jun, 2024
PATE-GAN复制的艰难追求：基准测试、审计、调试
The Elusive Pursuit of Replicating PATE-GAN: Benchmarking, Auditing, Debugging
HTML
PDF
Georgi Ganev, Meenatchi Sundaram Muthu Selva Annamalai, Emiliano De Cristofaro
TL;DR
分析和评估了六种开源的 PATE-GAN 实现，发现实现中存在架构偏差，无法复制原始论文的效用性能；在深度隐私评估中，发现所有实现泄露了比预期更多的隐私，并揭示了17个隐私违规和5个其他错误。
Abstract
Synthetic data created by differentially private (DP) generative models is increasingly used in real-world settings. In this context,
pate-gan
has emerged as a popular algorithm, combining
generative adversarial network
→