New methods enable efficient exact unlearning for federated AI training

Researchers have developed new methods for efficient exact unlearning in federated zeroth-order optimization (FedZOO). These techniques, FedZOUL-S and FedZOUL-NS, leverage total variation stability to reuse computation and reduce the cost of removing specific training data's influence. Experiments show these methods achieve competitive model utility with significantly less unlearning time compared to re-optimization, and drastically lower communication overhead than first-order baselines. AI

IMPACT Enables more robust and privacy-preserving collaborative AI training by efficiently handling data deletion requests.

RANK_REASON Academic paper detailing new algorithms and their theoretical guarantees. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• arXiv
• Federated zeroth-order optimization
• FedZOO
• FedZOUL-NS
• FedZOUL-S
• Gaussian noise
• Hugging Face
• machine learning

• paper
• model release
• infra

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