New FedLNS framework combats adversarial attacks in federated LLMs

Researchers have developed FedLNS, a new server-side framework designed to protect federated learning models from adversarial manipulation. This method uses the signature of changes in normalization-layer parameters within client updates to identify and screen out malicious contributions. FedLNS requires no additional data exchange from clients and can be integrated with standard federated learning aggregation techniques. Experiments demonstrate its effectiveness in reducing test perplexity for GPT-style, BERT-style, and LLaMA-style models, even with up to 40% of clients engaging in target manipulation. AI

IMPACT Enhances the security and reliability of federated learning for LLMs, enabling more robust private data training.

RANK_REASON The cluster contains an academic paper detailing a new method for mitigating adversarial manipulation in federated LLMs. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Bert
• federated learning
• FedLNS
• generative pre-trained transformer
• LayerNorm
• llama

• paper
• safety

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