CARD: Closed-loop Adversarial Robustness Distillation Framework
Start here
An arXiv paper released on October 14, 2026 proposes a closed-loop adversarial robustness distillation framework called CARD: first perform strong-to-weak adversarial distillation, then use reliability-enhanced student feedback to refine the robust teacher, and feed the teacher's gains back to the same student. The method combines reverse adversarial student feedback with teacher-student difference maximization, and provides sufficient conditions for reliable weak-to-strong feedback transfer. On CIFAR-10, CIFAR-100, and TinyImageNet-10, CARD improves adversarial robustness in most settings and largely maintains clean accuracy.
Generated by AI from reports · Updated 29 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
- CARD: Closing the Teacher-Student Loop for Adversarial Robustness Distillation
arXiv · 2026-10-14 · Deep reading
Progress timeline
- arXiv: Adversarial Attacks, Backdoors & PoisoningCARD: Closing the Teacher-Student Loop for Adversarial Robustness Distillation
The researchers propose a closed-loop adversarial robustness distillation framework, CARD: first strong-to-weak adversarial distillation, then refining the robust teacher with reliability-enhanced student feedback, and passing the teacher's gains back to the same student. The method combines reverse adversarial student feedback with teacher-student divergence maximization, and provides sufficient conditions for reliable weak-to-strong feedback transfer. On CIFAR-10, CIFAR-100, and TinyImageNet-10, CARD improves adversarial robustness in most settings while largely maintaining clean accuracy.