StratFL: Defending Against Stealthy Model Poisoning Attacks in Federated Learning
Start here
An arXiv paper released on October 16, 2026 proposes StratFL, which models the attack and defense of covert model poisoning (CMP) in federated learning as an incomplete repeated game in which neither side can observe the other's actions, and designs an adversarial search algorithm as well as belief update algorithms on both the server and attacker sides. Experiments on MNIST and CIFAR-10 show that its accuracy is 10% higher than the current best method, and the belief update algorithm converges in multiple security scenarios.
Generated by AI from reports · Updated 25 days ago
Attention and developments
Attention, hourly
Now 20, 0% in 24 hours·Peak 20 (Nov 6 03:00)
Related research
This event's paper
- StratFL: Countering Covert Model Poisoning Attacks with Strategic Federated Learning
arXiv · 2026-10-15 · Deep reading
Progress timeline
- arXiv: Adversarial Attacks, Backdoors & PoisoningStratFL: Countering Covert Model Poisoning Attacks with Strategic Federated Learning
StratFL models the attack and defense of covert model poisoning (CMP) attacks in federated learning as an incomplete repeated game in which neither side can observe the other's actions, and designs an adversarial search algorithm along with belief update algorithms on both the server and attacker sides. Experiments on MNIST and CIFAR-10 show that its accuracy is 10% higher than the current best method, and the belief update algorithm converges in multiple security scenarios.