Hang Su

Xingxing Wei, Shouwei Ruan, Yinpeng Dong et al.

Adversarial patch is one of the important forms of performing adversarial attacks in the physical world. To improve the naturalness and aggressiveness of existing adversarial patches, location-aware patches are proposed, where the patch's location on the target object is integrated into the optimization process to perform attacks. Although it is effective, efficiently finding the optimal location for placing the patches is challenging, especially under the black-box attack settings. In this paper, we propose the Distribution-Optimized Adversarial Patch (DOPatch), a novel method that optimizes a multimodal distribution of adversarial locations instead of individual ones. DOPatch has several benefits: Firstly, we find that the locations' distributions across different models are pretty similar, and thus we can achieve efficient query-based attacks to unseen models using a distributional prior optimized on a surrogate model. Secondly, DOPatch can generate diverse adversarial samples by characterizing the distribution of adversarial locations. Thus we can improve the model's robustness to location-aware patches via carefully designed Distributional-Modeling Adversarial Training (DOP-DMAT). We evaluate DOPatch on various face recognition and image recognition tasks and demonstrate its superiority and efficiency over existing methods. We also conduct extensive ablation studies and analyses to validate the effectiveness of our method and provide insights into the distribution of adversarial locations.

Qiuju Yang, Hang Su, Lili Liu et al.

Auroral classification plays a crucial role in polar research. However, current auroral classification studies are predominantly based on images taken at a single wavelength, typically 557.7 nm. Images obtained at other wavelengths have been comparatively overlooked, and the integration of information from multiple wavelengths remains an underexplored area. This limitation results in low classification rates for complex auroral patterns. Furthermore, these studies, whether employing traditional machine learning or deep learning approaches, have not achieved a satisfactory trade-off between accuracy and speed. To address these challenges, this paper proposes a lightweight auroral multi-wavelength fusion classification network, MLCNet, based on a multi-view approach. Firstly, we develop a lightweight feature extraction backbone, called LCTNet, to improve the classification rate and cope with the increasing amount of auroral observation data. Secondly, considering the existence of multi-scale spatial structures in auroras, we design a novel multi-scale reconstructed feature module named MSRM. Finally, to highlight the discriminative information between auroral classes, we propose a lightweight attention feature enhancement module called LAFE. The proposed method is validated using observational data from the Arctic Yellow River Station during 2003-2004. Experimental results demonstrate that the fusion of multi-wavelength information effectively improves the auroral classification performance. In particular, our approach achieves state-of-the-art classification accuracy compared to previous auroral classification studies, and superior results in terms of accuracy and computational efficiency compared to existing multi-view methods.

Siyuan Zhang, Yichi Zhang, Yinpeng Dong et al.

Large Language Models (LLMs) often struggle to align their responses with objective facts, resulting in the issue of factual hallucinations, which can be difficult to detect and mislead users without relevant knowledge. Although post-training techniques have been employed to mitigate the issue, existing methods usually suffer from poor generalization and trade-offs in other different capabilities. In this paper, we propose to address these by directly augmenting LLM's fundamental ability to precisely leverage its knowledge and introduce PKUE (Precise Knowledge Utilization Enhancement), which fine-tunes the model on self-generated responses to precise and simple factual questions through preference optimization. Furthermore, we construct FactualBench, a comprehensive and precise factual QA dataset containing 181k Chinese data spanning 21 domains, to facilitate both evaluation and training. Extensive experiments demonstrate that PKUE significantly improves LLM overall performance, with consistent enhancement across factual tasks of various forms, general tasks beyond factuality, and tasks in different language.

Jiawei Chen, Zhengwei Fang, Xiao Yang et al.

Ensuring the safety and alignment of Large Language Models is a significant challenge with their growing integration into critical applications and societal functions. While prior research has primarily focused on jailbreak attacks, less attention has been given to non-adversarial failures that subtly emerge during benign interactions. We introduce secondary risks a novel class of failure modes marked by harmful or misleading behaviors during benign prompts. Unlike adversarial attacks, these risks stem from imperfect generalization and often evade standard safety mechanisms. To enable systematic evaluation, we introduce two risk primitives verbose response and speculative advice that capture the core failure patterns. Building on these definitions, we propose SecLens, a black-box, multi-objective search framework that efficiently elicits secondary risk behaviors by optimizing task relevance, risk activation, and linguistic plausibility. To support reproducible evaluation, we release SecRiskBench, a benchmark dataset of 650 prompts covering eight diverse real-world risk categories. Experimental results from extensive evaluations on 16 popular models demonstrate that secondary risks are widespread, transferable across models, and modality independent, emphasizing the urgent need for enhanced safety mechanisms to address benign yet harmful LLM behaviors in real-world deployments.