EventPast event
GCD: Gradient Confusion Defends Against Black-Box Query Attacks
Start here
AI overview
An arXiv paper proposes Gradient Confusion Defense (GCD), a deterministic defense that requires no retraining, creating gradient confusion based on the model Jacobian to increase the query complexity of score-based black-box adversarial attacks. On CIFAR-10 and ImageNet, the method makes attacks against models such as ResNet require more queries.
Generated by AI from reports · Updated 24 days ago
Attention and developments
Attention, hourly
Now 23, 0% in 24 hours·Peak 23 (Nov 6 03:00)
Nov 7Nov 8Nov 9Nov 10Nov 11Nov 12Now 23
Related research
This event's paper
- Mitigating Black-Box Score-Based Query Attacks via Gradient Confusion
arXiv · 2026-10-12 · Deep reading
Progress timeline
Oct 13
- arXiv: Adversarial Attacks, Backdoors & PoisoningGradient Confusion Defense (GCD): Mitigating Score-Based Black-Box Query Attacks via Gradient Confusion
Gradient Confusion Defense (GCD) is a retraining-free deterministic defense that increases the query complexity of score-based black-box adversarial attacks. By creating confusion based on the model Jacobian, it achieves on CIFAR-10/ImageNet with Res