New GDBR attack breaches privacy in federated learning with partial encryption

Researchers have developed a new attack called GDBR that can recover private labels from federated learning systems that use partial gradient encryption. This attack exploits a vulnerability in neural network architectures, demonstrating that encrypting only the output layer is insufficient for protecting sensitive data. GDBR can be used for downstream attacks like data reconstruction and membership inference, challenging the assumption that partial encryption provides adequate privacy. AI

IMPACT Highlights significant privacy risks in federated learning systems, potentially impacting the adoption of privacy-preserving AI techniques.

RANK_REASON Academic paper detailing a new attack method. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• federated learning
• machine learning
• Rui Zhang
• Transformer-based networks

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →