As language models are increasingly deployed in real-world scenarios with important outcomes, the impact of their mistakes often extends far beyond what correctness can capture. In domains like medicine, a single error can lead to serious consequences and put patient safety at risk. However, mainstream evaluation proto...
Qi-Feng Lei, Zhi-Yong Yang, Zi-Tai Wang et al.· 0 citations
Diffusion feature, i.e., intermediate activations extracted from diffusion model backbones, has emerged as a promising approach for unifying generation and vision understanding (discrimination). Despite the advancement of diffusion backbones into Diffusion Transformers (DiT), it has been found in previous literature th...
Ben-Yuan Meng, Yi-Liang Zhang, Jin-Wen Wu et al.· 0 citations
The authors' results provide a mechanistic explanation for several empirical patterns: squared-hinge loss is often an effective AUC surrogate; cross-entropy warm-up can help models that enter the AUC phase from a poor starting region; and parameter-efficient fine-tuning from a pretrained model can reduce that need by s...
Jia Chen, Zhi-Yong Yang, Shi-Long Bao et al.· 0 citations
Sampling-based test-time scaling enhances LLM reasoning by aggregating multiple reasoning paths for the same query. Existing methods use answer frequencies or path confidence to compute candidate scores, which do not explicitly characterize comparisons between competing answers. Inspired by the Condorcet principle in s...
Cong Hua, Qian-Qian Xu, Zi-Tai Wang et al.· 0 citations
Explicit reasoning in large language models (LLMs) is often associated with reliability, yet it can still pose risks when a model infers harmful goals under a benign context. We identify a distinct challenge: ensuring the safety of the \emph{multi-step reasoning process} itself, where locally coherent steps may still l...
Hao Yang, Ke Ma, Yangbangyan Jiang et al.· 0 citations
The visual anchoring effect is revealed: by enforcing consistency, the mechanism prevents the generated content from drifting away from the original harmful intent, thereby eliminating the model's natural safety escape route from harmful to benign content.
Peng Li, Qian-Qian Xu, Yangbangyan Jiang et al.· 0 citations
The CVPR 2026@AdvML Workshop Challenge on adversarial multimodal attacks against autonomous-driving VLAs is presented, providing a practical reference for future robustness evaluation and defense design in multimodal autonomous-driving systems.
Tian-Yuan Zhang, Zonglei Jing, Jiangfan Liu et al.· arXiv.org· 0 citations
UniTraffic-Agent is introduced, the MR-CAS solution for Track~3 of the 10th AI City Challenge, which includes Traffic Anomaly Reasoning (TAR) and two out-of-domain evaluations: FETV for fisheye traffic events and PSI-VQA for pedestrian intention reasoning.
Peng Li, Qianqian Xu, Shilong Bao et al.· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
