Continuous Collision Detection is essential in simulation and modeling for accurately identifying object collisions. While robust CCD techniques have matured for triangle meshes, ensuring floating-point robustness for parametric surfaces remains an open challenge due to their representational complexity and heightened algorithmic sensitivity. In this paper, we present the first floating-point-robust CCD framework for parametric surfaces. Built on the Time-Dependent Inclusion-Based Method (TDIBM), our approach introduces a novel error decomposition strategy that separates coefficient and arithmetic errors, enabling structured analysis and safety guarantees. To rigorously benchmark robustness, we develop a rational-arithmetic-based dataset by inverting the CCD process: we generate exact ground-truth datasets from prescribed collision outcomes. Our construction captures both typical scenarios and near-degenerate cases. We evaluate several CCD algorithms using this benchmark to provide an in-depth analysis. Together, our method and dataset establish a comprehensive foundation for analyzing, benchmarking, and improving floating-point robustness in parametric surface CCD. Code and dataset will be published upon acceptance.
论文检索
输入标题、作者或关键词，从 3,443 篇学术成果中精准定位
Cultural context profoundly shapes how people interpret online content, yet vision–language models (VLMs) remain predominantly trained through Western or English-centric lenses. This limits their fairness and cross-cultural robustness in tasks like hateful meme detection. We introduce a systematic evaluation framework designed to diagnose and quantify the cross-cultural robustness of state-of-the-art VLMs across multilingual meme datasets, analyzing three axes: (i) learning strategy (zero-shot vs. one-shot), (ii) prompting language (native vs. English), and (iii) translation effects on meaning and detection. Results show that the common ''translate-then-detect'' approach deteriorate performance, while culturally aligned interventions — native-language prompting and one-shot learning — significantly enhance detection. Our findings reveal systematic convergence toward Western safety norms and provide actionable strategies to mitigate such bias, guiding the design of globally robust multimodal moderation systems.
Mind the Ambiguity: Aleatoric Uncertainty Quantification in LLMs for Safe Medical Question Answering
The deployment of Large Language Models in Medical Question Answering is severely hampered by ambiguous user queries, a significant safety risk that demonstrably reduces answer accuracy in high-stakes healthcare settings. In this paper, we formalize this challenge by linking input ambiguity to aleatoric uncertainty (AU), which is the irreducible uncertainty arising from underspecified input. To facilitate research in this direction, we construct CV-MedBench, the first benchmark designed for studying input ambiguity in Medical QA. Using this benchmark, we analyze AU from a representation engineering perspective, revealing that AU is linearly encoded in LLM's internal activation patterns. Leveraging this insight, we introduce a novel AU-guided ''Clarify-Before-Answer'' framework, which incorporates AU-Probe - a lightweight module that detects input ambiguity directly from hidden states. Unlike existing uncertainty estimation methods, AU-Probe requires neither LLM fine-tuning nor multiple forward passes, enabling an efficient mechanism to proactively request user clarification and significantly enhance safety. Extensive experiments across four open LLMs demonstrate the effectiveness of our QA framework, with an average accuracy improvement of 9.48% over baselines. Our framework provides an efficient and robust solution for safe Medical QA, strengthening the reliability of health-related applications. The code is available at https://github.com/yaokunliu/AU-Med.git, and the CV-MedBench dataset is released on Hugging Face at https://huggingface.co/datasets/yaokunl/CV-MedBench.
As web platforms evolve towards greater personalization and emotional complexity, conversational agents must transcend superficial empathy to demonstrate identity-aware emotional reasoning. However, existing systems face two limitations: (1) reliance on situation-centric datasets lacking persistent user identity, which hampers the capture of personalized affective nuances; and (2) dependence on opaque, coarse reward signals that hinder development of verifiable empathetic reasoning. To address these gaps, we introduce KardiaBench, a large-scale user-grounded benchmark comprising 178,080 QA pairs across 22,080 multi-turn conversations anchored to 671 real-world profiles. The dataset is constructed via a model-in-the-loop pipeline with iterative rubric-guided refinement to ensure psychological plausibility and persona consistency. This progressive empathy pipeline that integrates user comprehension, contextual reasoning, and emotion perception into conversations, followed by iterative critique and rubric-based refinement to ensure psychological plausibility, emotional fidelity, and persona consistency. Building on this, we propose Kardia-R1, a framework that trains models for interpretable, stepwise empathetic cognition. Kardia-R1 leverages Rubric-as-Judge Empathetic Reinforcement Learning (Rubric-ERL), a GRPO-based method that uses explainable, human-aligned rubric rewards to tightly couple user understanding, emotional inference, and supportive response generation. Extensive experiments across four LLM backbones demonstrate that Kardia-R1 consistently outperforms other methods in emotion accuracy, empathy, relevance, persona consistency, and safety.
Large language models (LLMs) are increasingly deployed in domains where errors carry high social, scientific, or safety costs. Yet standard confidence estimators, such as token likelihood, semantic similarity and multi-sample consistency, remain brittle under distribution shift, domain-specialised text, and compute limits. In this work, we present Structural Confidence, a single-pass, model-agnostic framework that enhances output correctness prediction based on multi-scale structural signals derived from a model's final-layer hidden-state trajectory. By combining spectral, local-variation, and global shape descriptors, our method captures internal stability patterns that are missed by probabilities and sentence embeddings. We conduct extensive, cross-domain evaluation across four heterogeneous benchmarks—FEVER (fact verification), SciFact (scientific claims), WikiBio-hallucination (biographical consistency), and TruthfulQA (truthfulness-oriented QA). Our Structural Confidence framework demonstrates strong performance compared with established baselines in terms of AUROC and AUPR. More importantly, unlike sampling-based consistency methods which require multiple stochastic generations and an auxiliary model, our approach uses a single deterministic forward pass, offering a practical basis for efficient, robust post-hoc confidence estimation in socially impactful, resource-constrained LLM applications.
In the rapidly evolving landscape of e-commerce, visual search has become a cornerstone of user experience, enabling customers to find products using images rather than traditional text queries. However, a comprehensive analysis reveals a persistent challenge: nearly half of retrieval failures stem from systems that prioritize superficial visual similarity over semantic relevance, resulting in frustrating user experiences where searches return visually similar but functionally different products. This limitation becomes particularly acute in Business-to-Business environments, where incorrect product recommendations can have significant operational and safety implications. In this paper, we propose a novel solution, Adaptive Retrieval with Category-aware Hierarchical sEmantic Refinement (ARCHER), which presents a novel multimodal retrieval framework that addresses these challenges through progressive semantic alignment. Unlike existing approaches that treat all visual similarities equally, ARCHER employs a sophisticated three-stage learning strategy that systematically builds from coarse-grained category understanding to fine-grained product discrimination. The framework begins with Proto-Align Enhancement to establish foundational visual-textual correspondences, progresses through Cross-View Learning to develop robust viewpoint-invariant representations, and culminates with Margin-based Representation Enhancement that learns to distinguish between visually similar but functionally distinct products. Most significantly, the framework has been successfully deployed on Alibaba.com's B2B platform since December 2024, where it serves millions of daily queries and has achieved a measurable 2.1% improvement in click-through rates.
Using Web Mining and Content Analysis to find and understand clues from the massive amount of unstructured text online is very important for predicting future events and providing early risk warnings in important fields like finance and public safety. While Large Language Models (LLMs) exhibit potential in processing and understanding text, current text-based event prediction faces two primary challenges: first, an insufficient utilization of potential information within the text, such as causal relationships and latent associations, and second, limited predictive reliability constrained by issues like the LLM's own ability and hallucinations. To address these challenges, we propose a novel event prediction framework, Bidirectional Reasoning with Self-Correction (BRSC). BRSC comprises two complementary reasoning dimensions: temporal deductive reasoning, which analyzes the trajectory of historical events along the timeline to enable accurate trend extrapolation, and synchronic associative reasoning, which deeply mines details and latent connections from documents within a specific time window to extended semantic information. In addition, we use a self-correction mechanism that identifies and rectifies potential hallucinations and errors during the reasoning process. Extensive experiments on international relations event prediction demonstrate that BRSC achieves significant improvements over several leading LLM-based methods.
As critical infrastructure for Web 3.0, DAG-based blockchains promise high throughput for DeFi, IoT, and DApps. However, realizing this potential is challenging, as system performance is dictated by a multitude of interdependent parameters across network, node, and consensus layers. Manual configuration fails to adapt to dynamic workloads, leading to suboptimal performance. We introduce Alzo, a novel auto-tuner that employs hierarchical reinforcement learning (HRL) to navigate this complex configuration space. By decomposing the DAG blockchain's workflow into distinct stages, Alzo's HRL policy learns from stage-level performance metrics to control critical parameters governing consensus, execution, and graph topology in real-time. Furthermore, we employ a shadow-control loop to ensure the safety of all parameter adjustments. Our experiments show that Alzo significantly outperforms other configurations, achieving higher throughput and lower latency under variable workloads with minimal overhead.
Online hate on social media ranges from overt slurs and threats (hard hate speech ) to soft hate speech: discourse that appears reasonable on the surface but uses framing and value-based arguments to steer audiences toward blaming or excluding a target group. We hypothesize that current moderation systems, largely optimized for surface toxicity cues, are not robust to this reasoning-driven hostility, yet existing benchmarks do not measure this gap systematically. We introduce SoftHateBench, a generative benchmark that produces soft-hate variants while preserving the underlying hostile standpoint. To generate soft hate, we integrate the Argumentum Model of Topics (AMT) and Relevance Theory (RT) in a unified framework: AMT provides the backbone argument structure for rewriting an explicit hateful standpoint into a seemingly neutral discussion while preserving the stance, and RT guides generation to keep the AMT chain logically coherent. The benchmark spans 7 sociocultural domains and 28 target groups, comprising 4,745 soft-hate instances. Evaluations across encoder-based detectors, general-purpose LLMs, and safety models show a consistent drop from hard to soft tiers: systems that detect explicit hostility often fail when the same stance is conveyed through subtle, reasoning-based language. Disclaimer. Contains offensive examples used solely for research.
Dynamic graph clustering aims to detect and track time-varying clusters in dynamic graphs, revealing how complex real-world systems evolve over time. However, existing methods are predominantly black-box models. They lack interpretability in their clustering decisions and fail to provide semantic explanations of why clusters form or how they evolve, severely limiting their use in safety-critical domains such as healthcare or transportation. To address these limitations, we propose an end-to-end interpretable framework that maps continuous graph embeddings into discrete semantic concepts through learnable prototypes. Specifically, we first decompose node representations into orthogonal role and clustering subspaces, so that nodes with similar roles (e.g., hubs, bridges) but different cluster affiliations can be properly distinguished. We then introduce five node role prototypes (Leader, Contributor, Wanderer, Connector, Newcomer) in the role subspace as semantic anchors, transforming continuous embeddings into discrete concepts to facilitate LLM understanding of node roles within communities. Finally, we design a hierarchical LLM reasoning mechanism to generate both clustering results and natural language explanations, while providing consistency feedback as weak supervision to refine node representations. Experimental results on four synthetic and six real-world benchmarks demonstrate the effectiveness, interpretability, and robustness of DyG-RoLLM. Code is available at https: //github.com/Clearloveyuan/DyG-RoLLM.
With the growing importance of content safety, toxic language detection, especially in Chinese online environments, has become a key task in natural language processing. However, real-world toxic expressions often appear in obfuscated forms such as pinyin abbreviations, symbol insertion, or visually similar substitutions, making them difficult to detect using traditional rule-based or static models. To address this challenge, we propose a dynamic knowledge graph construction method for toxic text variants, named Variant-KG. This graph encodes diverse structural relations between canonical toxic terms and their variants based on phonetic similarity, visual resemblance, and contextual co-occurrence. A small amount of labeled data is further used to fine-tune large language models (LLMs), enabling initial normalization and variant recognition. On top of this, we design a collaborative detection framework that combines the Variant-KG with frozen LLMs. It performs graph augmented prompting for structure-aware reasoning and adopts a Think-Search-Generate paradigm to dynamically recover broken paths when graph connections are incomplete, enabling both data self-enhancement and knowledge completion during inference. Evaluations on multiple Chinese toxic language datasets show that our model consistently outperforms both non-knowledge-enhanced and existing knowledge-enhanced baselines, demonstrating the effectiveness of our proposed dynamic reasoning framework in handling diverse toxic expressions.
The rapid proliferation of large language models (LLMs) in web-scale applications has heightened the need for robust security mechanisms. To address these challenges, various guardrail strategies have been introduced, including input filters, output moderation, and model alignment. Among them, input-side guardrails are commonly adopted as the first line of defense against adversarial prompts. However, we identify a critical architectural weakness: a representation asymmetry between input-side guardrails and the core LLMs, which creates a security gap exploitable by adversaries. Through systematic analysis, we show that guardrails are fragile when facing minor character perturbations, while LLMs remain semantically resilient and can still reconstruct malicious intent from noisy inputs. Exploiting this asymmetry, we propose RepMism, a hybrid adversarial framework that combines character injection with chain-of-thought hijacking, coordinated through hierarchical scheduling and safety continuation. Our extensive empirical evaluation across four commercial LLMs and five state-of-the-art guardrails demonstrates RepMism's strong attack capability, achieving high success rates. Importantly, we identify a ''success interval'' where perturbations effectively bypass guardrail detection yet remain interpretable to target models. These findings expose key flaws in current multi-layer security architectures and offer actionable guidance for building more resilient defenses through perturbation-aware training and cross-layer representation alignment.
Large Language Models (LLMs) have achieved remarkable progress in understanding and generation tasks, yet they remain highly susceptible to adversarial prompt attacks that bypass safety safeguards and induce the generation of harmful content. Existing pre-generation and post-generation defense methods typically rely on intent recognition, surface-level features, or external classifiers, rendering them vulnerable to evasion via subtle prompt perturbations while incurring substantial computational overhead. In this paper, we propose PADD (Prefix-based Attention Divergence Detector), a lightweight pre-generation defense mechanism that leverages internal model signals for robust attack detection. At its core, PADD prepends a lightweight safety prefix to the input prompt and compares attention distributions between the original and prefixed prompts. By transforming cross-prompt comparisons into self-comparisons via composite signals of attention divergence and attention plasticity, PADD achieves strong separability between adversarial and benign prompts with low-latency detection, without requiring modifications or fine-tuning of the base model. Extensive experiments across four mainstream open-source LLMs and multiple public benchmarks demonstrate that PADD significantly reduces attack success rates (0.4--3.0%) while maintaining low false rejection rates (0--5.2%). These results position PADD as a scalable, efficient, and practical safeguard for LLM safety.
The InterPlanetary File System (IPFS) is the largest decentralized content-centric storage network. While its architecture enables resilient, distributed content delivery, it can be abused to host and disseminate malicious content. Public IPFS HTTP gateways further expand this threat surface, enabling attackers to deploy phishing websites and leverage gateway reputation to evade detection. This model can keep content available even after attackers go offline and challenges traditional phishing detection systems. We present a framework for monitoring and characterizing phishing on IPFS, leveraging a measurement platform that integrates multi-source data, including IPFS traffic and passive DNS. Over 11 months, we detect 10,489 phishing CIDs, grouped into 448 phishing clusters. 80% of detected CIDs originate from only 69 clustered campaigns indicating that targeting a small number of dominant clusters could yield high mitigation leverage. We also identify 588 gateways involved in dissemination, including 573 outside public gateway lists, and show that attackers can exploit caching across reputable gateways to amplify attacks and extend content availability. Finally, we find that traditional Web phishing countermeasures and IPFS blocklists provide insufficient protection. Our findings support practical mitigation and offer broader insights for trust and safety in decentralized web infrastructures.
The effectiveness upper bound of retrieval-augmented generation (RAG) is fundamentally constrained by the semantic integrity and information granularity of text chunks in its knowledge base. Moreover, domain documents are characterized by dense terminology and strong contextual dependencies, which exacerbate the semantic fragmentation of text chunks, thereby making it difficult to efficiently utilize their key information. To address these challenges, this paper proposes QChunker, which restructures the RAG paradigm from retrieval-augmentation to understanding-retrieval-augmentation. Firstly, QChunker models the text chunking as a composite task of text segmentation and knowledge completion to ensure the logical coherence and integrity of text chunks. Drawing inspiration from Hal Gregersen's ''Questions Are the Answer'' theory, we design a multi-agent debate framework comprising four specialized components: a question outline generator, text segmenter, integrity reviewer, and knowledge completer. This framework operates on the principle that questions serve as catalysts for profound insights. Through this pipeline, we successfully construct a high-quality dataset of 45K entries and transfer this capability to small language models. Additionally, to handle long evaluation chains and low efficiency in existing chunking evaluation methods, which overly rely on downstream QA tasks, we introduce a novel direct evaluation metric, ChunkScore. Both theoretical and experimental validations demonstrate that ChunkScore can directly and efficiently discriminate the quality of text chunks. Furthermore, during the text segmentation phase, we utilize document outlines for multi-path sampling to generate multiple candidate chunks and select the optimal solution employing ChunkScore. Extensive experimental results across four heterogeneous domains exhibit that QChunker effectively resolves aforementioned issues by providing RAG with more logically coherent and information-rich text chunks. Notably, this study also establishes a small-domain QA dataset concerning hazardous chemical safety, which fully reveals the significant value of RAG in specialized domains and the generalization capability of the QChunker framework.
Intelligent agents operating in biomedical laboratories need physics-aware understanding beyond simple geometry or semantics. In-the-wild web images capture authentic lab interactions, yet methods designed for instrumented 3D capture and reconstruction struggle to turn this resource into functional knowledge. Consequently, current autonomous laboratory systems lack a queryable representation of affordances for precise object and liquid handling tasks. We pose a web-native question: how to convert single-view, uncalibrated web images into a structured, physics-aware scene representation. We introduce WPIS (Web- and Physics-Informed Scene-understanding), a pipeline that compiles Physics-aware 3D Scene Graphs (P-3DSGs) from web imagery by fusing open-vocabulary instance/mask cues with relative geometry, augmenting nodes with real-valued liquid states and fine-grained hand–object interaction (HOI) subgraphs, and attaching concise natural-language functional relations—without intrinsics, multi-view, or CAD priors. We release WebLab-3DSG, a 1,000-scene knowledge base pairing each RGB image with its P-3DSG JSON, relative depth, and a single-image point-cloud proxy. In an expert study, grounding an LLM in a P-3DSG improves answer quality by 40% over a strong VLM that reasons directly from RGB, with the largest gains on feasibility, HOI alignment, and safety constraints. WPIS offers a reproducible path to physics-aware reasoning from in-the-wild web imagery and a practical substrate for decision-making in autonomous lab settings.
Graph Neural Networks (GNNs) are vulnerable to graph backdoor attacks, which poses severe risks to their deployment in safety-critical applications. Existing defenses predominantly focus on specific backdoor triggers, making them brittle and unable to generalize across different backdoor triggers with varying properties. Motivated by this limitation, this work proposes an attack-agnostic graph backdoor defense mechanism A2GBD, which does not require prior knowledge of the specific attack strategies (e.g., edge perturbation, node attribute manipulation) to achieve effective defense. A2GBD consists of suspicious node selection and defense strategy generation. The selection module selects high-suspicion nodes to enhance defense awareness, while the defense agent adaptively determines and executes defense strategies. Extensive experiments on multiple benchmark datasets demonstrate that A2GBD consistently lowers attack success rates while maintaining high clean accuracy, showing strong robustness and generalizability against diverse graph backdoor attack strategies.
The canonical model for multi-channel marketing, Bandits with Knapsacks (BwK), optimizes cumulative rewards subject to resource constraints but typically assumes a fixed, exogenous budget. This assumption is tenuous in real-world systems requiring performance-adaptive investment, where pre-committing to a budget is challenging and suboptimal. We introduce Marketing Hosting, a paradigm modeling the budget as an endogenous, performance-dependent variable. This yields a new problem class, Bandits with Endogenous Knapsacks (BwEK), characterized by a challenging feedback loop coupling rewards with constraints. We develop a specialized primal-dual algorithm to manage this coupling. For settings with hard, per-round constraints, we design a novel risk-aware algorithm that mitigates the path-dependent risk of ruin, providing the first high-probability safety guarantee for such problems. Finally, we solve the strategic bi-level problem of learning the optimal reinvestment rate. We validate our theoretical results through extensive simulations, real-world data experiments, and a live A/B test.
The Digital Roads project at Cambridge University is leveraging digitalisation, automation, and low-carbon materials to build an Automated Maintenance Plant (AMP) for UK road networks, aimed at minimising repair times to reduce congestion, improving safety, and contributing to the UK’s net-zero goals through faster, more accurate, and efficient road maintenance.