Jiancheng Shi

2 accepted papers

2026

Posterior Mismatch Matters: Adversarial Training for Long-Tailed Robustness

Lilin Zhang, Li Yue, Jiancheng Shi, Jiancheng Lv, Xianggen Liu

ICML 2026poster

Adversarial training breaks down in long-tailed settings, exhibiting severe robustness degradation on worst-performing (often tail) classes. We identify a key cause of this failure as a posterior mismatch: coarse-grained absolute labels collapse class posteriors into point estimates, leading to bias…

Cited by 0 Source ↗Scholar ↗

2026

Taming the Long Tail: Rebalancing Adversarial Training via Adaptive Perturbation

Lilin Zhang, Yimo Guo, Yue Li, Jiancheng Shi, Xianggen Liu

CVPR 2026

Deep neural networks are highly vulnerable to adversarial examples, i.e.,small perturbations that can significantly degrade model performance. While adversarial training has become the primary defense strategy, most studies focus on balanced datasets, overlooking the challenges posed by real-world l

Cited by 0 Source ↗code ↗Scholar ↗