OPFL: Optimistic Verification of Federated Learning via Empirical Boundary
OPFL verifies federated learning inside MPC with an empirical gradient boundary, stopping model poisoning at much lower cost.
OPFL is an optimistic verification framework for privacy-preserving federated learning. Client steps are replayed inside secure multi-party computation, and an offline empirical boundary separates benign numeric differences between MPC and local GPUs from malicious updates. Only sampled steps are audited. On LeNet, BERT, and Qwen, the authors report 0% attack success against model poisoning and PGD attacks; at sampling probability 0.01, LeNet verification was about 98.6 times faster than full MPC-based FL and 625.5 times faster than a zero-knowledge approach.
- OPFL replays sampled client training steps inside secure multi-party computation.
- An offline empirical boundary separates benign GPU-MPC drift from malicious updates.
- LeNet, BERT, and Qwen tests report 0% success for poisoning and PGD attacks.
- At p=0.01, LeNet verification is about 98.6× faster than full MPC FL.
From the original article
1 min read at arxiv.orgFederated learning enables multiple clients to collaboratively train models without sharing their private data. However, the lack of visibility into local training makes it difficult to verify whether clients follow the prescribed training procedure or submit malicious updates, such as model poisoning. A natural approach is to replay client training for verification.
By Hongxu Su · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.