New CACTUS method implants semantic backdoors in decentralized federated learning

Researchers have developed CACTUS, a novel method for implanting semantic backdoors in decentralized federated learning systems. This technique converts label-consistent semantic pairs into target-directed representation shifts, isolating trigger effects and coupling them across samples. CACTUS demonstrated a mean attack success rate of 51.2% on the Speech Commands dataset and achieved the highest mean ASR on three out of four tested modalities under various aggregation rules, indicating its effectiveness in propagating backdoors through repeated decentralized aggregation. AI

IMPACT This research highlights potential security vulnerabilities in decentralized federated learning systems, prompting further investigation into robust defense mechanisms.

RANK_REASON The cluster contains a research paper detailing a new method for backdoor attacks in federated learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• alphaXiv
• arXiv
• CACTUS
• CatalyzeX
• CORE Recommender
• DagsHub
• Decentralized federated learning system
• federated learning
• Gotit.pub
• Hugging Face
• IArxiv Recommender
• Influence Flower
• ScienceCast
• Speech Commands

• paper
• safety

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