Open knowledge bases (e.g., websites) are widely adopted in Retrieval-Augmented Generation (RAG) systems to provide supplementary knowledge (e.g., latest information). However, such sources inevitably contain biased or harmful content, and incorporating these untrusted contents into the RAG process introduces significant safety risks, including the degradation of LLM performance and the potential generation of harmful outputs. Recent studies have shown that this vulnerability can be further amplified by adversarial poisoning attacks specifically targeting the knowledge sources. Most existing methods primarily emphasize improving the accuracy and efficiency of RAG systems, usually overlooking these critical safety concerns. In this paper, we propose a safety-aware retrieval framework (ShieldRAG) designed to augment language model generation by jointly optimizing for both relevance and safety in the retrieved knowledge content. The core idea of ShieldRAG is to transfer the safety knowledge implicitly encoded in powerful LLMs into the retriever model through an adversarial knowledge alignment mechanism. This can empower the retriever with the safety awareness, and adapt to the diverse and unknown distribution of unsafe content encountered in practical scenarios. We evaluate ShieldRAG on seven real-world datasets using five widely-used LLMs and two state-of-the-art poisoning attack strategies. Experimental results show that our method substantially improves the robustness of RAG systems against unsafe knowledge sources, while maintaining competitive performance in terms of generation accuracy and efficiency.

Yuanhong Huang, Huili Wang, Xueying Bai, Jinrui Wang, Jiajun Liu, Ziqin Wang, Wanchun Ni, Shangguang Wang, Tao Qi. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Jinrui Wang, Zhenfeng Gao, Wendan Wang, Huili Wang, Zichen Qin, Linjie Zhu, Hongke Fu, Shangguang Wang, Tao Qi. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

This paper presents Visualization Badges, graphical labels shown alongside visualizations to communicate provenance and design considerations to enhance understandability and transparency. Badges may, for example, highlight a major finding, disclose that an axis has been truncated, or warn of possible visual artifacts. Inspired by nutrition and energy labels on product packaging, visualization badges aim (i) to allow visualization authors to justify and disclose analysis and design decisions and (ii) to make readers aware of important information when viewing and interpreting visualizations. Collectively, visualization badges aim to foster trust in visualizations and prevent readers from drawing incorrect conclusions. Based on a series of co-design workshops, we define and evaluate the concept of visualization badges and formulate a conceptual framework for analysis, application, and further research. Our framework includes a catalog of 132 visualization badges, categorization schemes, design options for their visual representations, applied visualization examples, and guidelines for their use. We hope that visualization badges will help communicate data and collectively improve communication, visualization literacy, and the quality of visualization techniques. Our badges, workshops, and guidelines can be found online https://vis-badges.github.io.

In the practical application of rotating machinery , the change of working conditions can meet different manufacturing requirements. When fault diagnosis is performed on monitoring data with different working conditions, the change of data distribution will bring interference information which is highly related to working conditions and inconsistent matching problems in the process of multi-target domain transfer. In order to solve these problems, a working condition decoupling adversarial network (WCDAN) is proposed for multi-target domain fault diagnosis. Specifically, the prototype discrepancy alignment module is constructed following a weight-shared wavelet convolution feature extractor to ensure a clear prototype representation boundary. Then, the adaptive domain discriminator weight, along with the acquired multi-domain discrepancy, are utilized to decouple the working conditions. This process filters out interference information that highly associated with the source domain working conditions while preserving the inherent fault characteristics. Furthermore, the strategy of multi-domain hybrid alignment aims to minimize the disparity between different domains and solve the inconsistent matching issue. Based on two gearbox fault datasets under stable and unstable conditions, the comparative experimental results show that the WCDAN can be generalized from a single source domain to multiple target domains at the same time and achieve excellent fault diagnosis performance.

This paper defines, analyzes, and discusses the emerging genre of visualization atlases. We currently witness an increase in web-based, data-driven initiatives that call themselves "atlases" while explaining complex, contemporary issues through data and visualizations: climate change, sustainability, AI, or cultural discoveries. To understand this emerging genre and inform their design, study, and authoring support, we conducted a systematic analysis of 33 visualization atlases and semi-structured interviews with eight visualization atlas creators. Based on our results, we contribute (1) a definition of a visualization atlas as a compendium of (web) pages aimed at explaining and supporting exploration of data about a dedicated topic through data, visualizations and narration. (2) a set of design patterns of 8 design dimensions, (3) insights into the atlas creation from interviews and (4) the definition of 5 visualization atlas genres. We found that visualization atlases are unique in the way they combine i) exploratory visualization, ii) narrative elements from data-driven storytelling and iii) structured navigation mechanisms. They target a wide range of audiences with different levels of domain knowledge, acting as tools for study, communication, and discovery. We conclude with a discussion of current design practices and emerging questions around the ethics and potential real-world impact of visualization atlases, aimed to inform the design and study of visualization atlases.

Research on the health diagnosis of mechanical equipment has developed unprecedentedly in recent years, and a large number of diagnostic solutions have considerably improved the stability of mechanical equipment in industrial production. However, such satisfactory diagnostic performance relies on a large number of data samples, which are frequently difficult to obtain in real industrial scenarios. The traditional strategy of data sharing is no longer advisable due to the potential conflict of interest among users. A federated transfer learning scheme is proposed to alleviate the data island problem in industrial production while protecting data privacy. This solution adopts a distributed structure, which includes local model training and global model update. A differential training scheme is proposed to enhance the domain adaptability of the local model. The central server evaluates the contribution ability of each local model to the target task. It also weights and aggregates each client model on the basis of parameter importance ranking in the form of model fusion. The target task of the experiment is performed on two sets of bearing datasets. By comparing with other diagnostic methods, a conclusion can be drawn that the proposed scheme provides a promising federated learning method while protecting client data privacy.

This paper addresses the emerging task of recognizing multiple retinal diseases from wide-field (WF) and ultra-wide-field (UWF) fundus images. For an effective use of existing large amount of labeled color fundus photo (CFP) data and the relatively small amount of WF and UWF data, we propose a supervised domain adaptation method named Cross-domain Collaborative Learning (CdCL). Inspired by the success of fixed-ratio based mixup in unsupervised domain adaptation, we re-purpose this strategy for the current task. Due to the intrinsic disparity between the field-of-view of CFP and WF/UWF images, a scale bias naturally exists in a mixup sample that the anatomic structure from a CFP image will be considerably larger than its WF/UWF counterpart. The CdCL method resolves the issue by Scale-bias Correction, which employs Transformers for producing scale-invariant features. As demonstrated by extensive experiments on multiple datasets covering both WF and UWF images, the proposed method compares favorably against a number of competitive baselines.

As a modern and international urban, the capital city of China, Beijing's developing has been seen by all over the world. Now the city's developing direction has been denied in the core aspects like residential, political, and culture by the central government, rather than the economic only in the traditional way. This paper is focusing on the research of the development of real estate in Beijing from 1990s, especially the areas of less developing area, which are the main obstacles of Beijing city's developing during the 11th five-year-planning. Using the questionnaire and GIS & RS spatial analysis method, we investigated more than 200 spots in the central city, the 8 districts of Beijing, to collect the field information of these areas. In most of these areas, there live the local people more than 100 years who are mostly the poor, it is the main task of harmony society's to resolve these person's living condition for improving their welfare. Following the new city planning instituted in 2005 by local government, we proposed several suggestions for these less developing areas, i.e. protecting the historical culture, mining the potential values, building new houses for them in the suburban, etc. and there must be other matched politics for reach these purposes.