SURA: Data-efficient backdoor detection based on forgetting in federated learning
Start here
On November 10, 2026, an arXiv paper proposed a federated learning backdoor detection method, SURA, which it says uses only 10 clean server images and outperforms other methods with the same data budget under six backdoor attacks; in repeated-training-mode loss-rise probing, it reuses the same proxy data to strengthen client responses, and for federated vision models containing batch normalization (BN), it only restores the pre-probe shared BN state at read time, eliminating drift.
Generated by AI from reports · Updated 2 days ago
Related research
This event's paper
- Data-Efficient Unlearning-Based Backdoor Detection in Federated Learning
arXiv · 2026-11-10 · Deep reading
Progress timeline
- arXiv: Adversarial Attacks, Backdoors & PoisoningSURA: Data-efficient backdoor detection based on forgetting in federated learning
Researchers propose SURA, a federated learning backdoor detection method that uses only 10 clean server images and outperforms other methods with the same data budget under six backdoor attacks. The method reuses the same small clean proxy data across multiple training-mode loss-increase probes, amplifies client responses, and for federated vision models containing batch normalization (BN) restores the pre-probe shared BN state only at read time, eliminating BN drift. Its robustness approaches the level of using a much larger proxy dataset.
Related discussion
Attention trend
New sources per day
- Nov 11: 1 new sources (1 outlets, 0 accounts)
- Nov 12: 0 new sources (0 outlets, 0 accounts)
- Nov 13: 0 new sources (0 outlets, 0 accounts)