Researchers have demonstrated a novel remote Rowhammer attack that targets the central server in federated learning systems. By manipulating clients to generate repetitive memory updates on the server, an adversarial agent, using reinforcement learning, can induce bit-flips in the server's dynamic random-access memory. This attack, showcased on systems performing speech recognition and CIFAR-10 classification, achieved a high repetition rate, potentially disrupting learning processes or leading to elevated privileges. AI
IMPACT Highlights a new class of security vulnerabilities in federated learning, potentially impacting the integrity and security of distributed AI training.
RANK_REASON Academic paper detailing a novel security vulnerability in federated learning systems. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- CIFAR-10
- dynamic random-access memory
- federated learning
- Jinsheng Yuan
- reinforcement learning
- row hammer
- speech recognition
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
