arXiv subjects

Lu Sun

Publications and source records attributed to Lu Sun.

At least 19 recordsLinked to original sources

arXiv2026

Ollama in the Wild: A Longitudinal Measurement of Exposed Ollama LLM Endpoints at Internet Scale

Self-hosted large language model (LLM) serving is emerging as a distinct category of Internet service, but we still know little about how these deployments appear and change on the public Internet. We present a 365-day longitudinal measurement of exposed Ollama endpoints (port 11434) from February 2025 to February 2026, combining daily active probing with GeoIP/ASN enrichment, PTR and port-443 host observations, and survival analysis. Across 362 observation days and approximately 4.8 million IP$\times$day observations, 26.4% of the 152,137 cumulative IPs appear for a single day; across five selected CVEs, only 0.43-2.90% of below-fix IPs upgraded in place; the top five countries/regions account for over 70% of weighted observations; and cloud and hosting providers dominate the top ASNs. These results characterize exposed Ollama as a structural exposure surface: persistent, growing, and heavily concentrated. At the same time, old versions, common model choices, cloud and hosting ASNs, PTR categories, and TLS certificate patterns remain visible across the year, indicating recurring insecure deployment practices in cloud infrastructure and the potential reach of provider-level mitigation.

cs.NI↗

arXiv2026

High-dimensional Supermode Photonics Enabled by Hierarchical Supersymmetric Transformation

Modes provide a fundamental degree of freedom for photonic information processing, yet conventional multimode waveguides exhibit non-equidistant effective-index distributions, making closely spaced modes vulnerable to intermodal crosstalk. Supermode photonics can overcome this limitation by geometrically engineering coupled waveguide arrays to realize large and equidistant effective-index spacing, but precise supermode excitation and detection remain challenging at the subwavelength scale. Here, we report a hierarchical second-order discrete supersymmetric (DSUSY) transformation method that enables high-purity excitation and extraction of arbitrary target supermodes in a compact and scalable architecture. We experimentally demonstrate six-supermode multiplexing systems on silicon-on-insulator and silicon nitride platforms. Benefiting from the large supermode index spacing and the isospectrality of DSUSY transformations, the fabricated devices exhibit low insertion losses (<2.6 dB) and intermodal crosstalk (<-11.1 dB) for all channels over a 100-nm wavelength range. A high-speed transmission experiment on the silicon device achieves an aggregate data rate of 1.2 Tbit/s, with all channel bit error rates below the 7% hard-decision forward-error-correction threshold. The method can further support polarization-insensitive architectures, enabling compact polarization-supermode hybrid multiplexing. This work provides a scalable route toward high-dimensional supermode photonics for high-capacity optical interconnects, highly parallel AI optical computing, and high-dimensional quantum information processing.

physics.optics↗

arXiv2026

The Benchmark Illusion: Pruned LLMs Can Pass Multiple Choice but Fail to Answer

Compressing large language models reduces memory use and inference cost, but it can also create failures that standard benchmarks miss. A pruned model may still perform well on multiple-choice evaluations, yet fail to answer the same question in open generation. We ask what pruning changes: does it erase the correct answer, or does it make the answer harder to produce as the top output? We study this question with multilingual question answering, tracking the same questions before and after pruning. We find a benchmark illusion. Under high-sparsity pruning, especially Wanda, models often fail in greedy open generation while still selecting the correct answer under multiple-choice scoring. In these recognition-only errors, the answer is usually not gone, but demoted: it often reappears with beam search, sampling, or one in-context example. Overall, multiple-choice benchmarks can overstate the usability of compressed LLMs, creating an evaluation blind spot. Compressed models should be tested on what they can produce, not only on what they can recognize.

cs.CL↗

arXiv2026

Integrated Supermode Photonics Enabled by Supersymmetric Transformation

We report a systematic methodology to obtain supermodes with equidistant effective index distribution and to excite arbitrary target supermodes with high precision. By employing a multi-well optical potential realized by a judiciously designed waveguide array, the supported supermodes achieve maximal spacing and an equidistant distribution in effective index. More importantly, we develop a 2nd-order discrete supersymmetric (DSUSY) transformation method that enables the excitation and detection of two supermodes at the same time and can be extended to any number of supermodes via simple cascading. Together, these findings overcome the long-standing bottlenecks in integrated supermode photonics and provide an intrinsically scalable route towards harnessing supermodes as a new degree of freedom for encoding, transmitting, and processing information. We experimentally demonstrate the feasibility and universality of this method by realizing two- and four-supermode multiplexing systems. Benefitting from the large effective index spacing between supermodes and the isospectral nature of the DSUSY transformation, the fabricated devices show low insertion losses (< 2.48 dB at 1550 nm) and intermodal crosstalk (< -18 dB at 1550 nm) for all mode channels over a 100-nm wavelength range (1500-1600 nm). The high-speed data transmission experiment performed on the four-channel system achieves an aggregate data rate of 1.024 Tb/s while maintaining considerably low bit error rates, underscoring the potential of supermode photonics for high-capacity on-chip optical communications. This work lays the foundation for integrated supermode photonics, which uses supermodes as a new degree of freedom for light manipulation and opens new avenues for supermode-based applications including but not limited to on-chip optical communications, intelligent optical computing and quantum information technologies.

physics.optics↗

arXiv2026

A$^3$: Towards Advertising Aesthetic Assessment

Advertising images significantly impact commercial conversion rates and brand equity, yet current evaluation methods rely on subjective judgments, lacking scalability, standardized criteria, and interpretability. To address these challenges, we present A^3 (Advertising Aesthetic Assessment), a comprehensive framework encompassing four components: a paradigm (A^3-Law), a dataset (A^3-Dataset), a multimodal large language model (A^3-Align), and a benchmark (A^3-Bench). Central to A^3 is a theory-driven paradigm, A^3-Law, comprising three hierarchical stages: (1) Perceptual Attention, evaluating perceptual image signals for their ability to attract attention; (2) Formal Interest, assessing formal composition of image color and spatial layout in evoking interest; and (3) Desire Impact, measuring desire evocation from images and their persuasive impact. Building on A^3-Law, we construct A^3-Dataset with 120K instruction-response pairs from 30K advertising images, each richly annotated with multi-dimensional labels and Chain-of-Thought (CoT) rationales. We further develop A^3-Align, trained under A^3-Law with CoT-guided learning on A^3-Dataset. Extensive experiments on A^3-Bench demonstrate that A^3-Align achieves superior alignment with A^3-Law compared to existing models, and this alignment generalizes well to quality advertisement selection and prescriptive advertisement critique, indicating its potential for broader deployment. Dataset, code, and models can be found at: https://github.com/euleryuan/A3-Align.

cs.CV↗

arXiv2026

aCAPTCHA: Verifying That an Entity Is a Capable Agent via Asymmetric Hardness

As autonomous AI agents increasingly populate the Internet, a novel security challenge arises: "Is this entity an AI agent?" It is a new entity-type verification problem with no established solution. We formalize the problem through a three-class entity taxonomy (Human, Script, Agent) based on a verifiable agentic capability vector (action, reasoning, and memory). A timing threshold t exploits the asymmetric hardness between human cognition and AI processing to separate the three classes. We define the Agentic Capability Verification Problem (ACVP) through three necessity primitives, each testing one capability dimension. Building on this foundation, we introduce aCAPTCHA (Agent CAPTCHA), a time-constrained security game for agent admission whose security rests on ACVP hardness under t. We instantiate aCAPTCHA through time-bounded natural-language understanding as a multi-round HTTP verification protocol, and evaluate it with preliminary agent trials that validate the protocol's soundness and completeness. aCAPTCHA provides a composable, infrastructure-free admission gate for any service where entity-type verification is required.

cs.CR↗

arXiv2026

SafeSci: Safety Evaluation of Large Language Models in Science Domains and Beyond

The success of large language models (LLMs) in scientific domains has heightened safety concerns, prompting numerous benchmarks to evaluate their scientific safety. Existing benchmarks often suffer from limited risk coverage and a reliance on subjective evaluation. To address these problems, we introduce SafeSci, a comprehensive framework for safety evaluation and enhancement in scientific contexts. SafeSci comprises SafeSciBench, a multi-disciplinary benchmark with 0.25M samples, and SafeSciTrain, a large-scale dataset containing 1.5M samples for safety enhancement. SafeSciBench distinguishes between safety knowledge and risk to cover extensive scopes and employs objective metrics such as deterministically answerable questions to mitigate evaluation bias. We evaluate 24 advanced LLMs, revealing critical vulnerabilities in current models. We also observe that LLMs exhibit varying degrees of excessive refusal behaviors on safety-related issues. For safety enhancement, we demonstrate that fine-tuning on SafeSciTrain significantly enhances the safety alignment of models. Finally, we argue that knowledge is a double-edged sword, and determining the safety of a scientific question should depend on specific context, rather than universally categorizing it as safe or unsafe. Our work provides both a diagnostic tool and a practical resource for building safer scientific AI systems.

cs.LG↗

arXiv2026

BLM-Guard: Explainable Multimodal Ad Moderation with Chain-of-Thought and Policy-Aligned Rewards

Short-video platforms now host vast multimodal ads whose deceptive visuals, speech and subtitles demand finer-grained, policy-driven moderation than community safety filters. We present BLM-Guard, a content-audit framework for commercial ads that fuses Chain-of-Thought reasoning with rule-based policy principles and a critic-guided reward. A rule-driven ICoT data-synthesis pipeline jump-starts training by generating structured scene descriptions, reasoning chains and labels, cutting annotation costs. Reinforcement learning then refines the model using a composite reward balancing causal coherence with policy adherence. A multitask architecture models intra-modal manipulations (e.g., exaggerated imagery) and cross-modal mismatches (e.g., subtitle-speech drift), boosting robustness. Experiments on real short-video ads show BLM-Guard surpasses strong baselines in accuracy, consistency and generalization.

cs.CV↗

arXiv2026

XMap: Fast Internet-wide IPv4 and IPv6 Network Scanner

XMap is an open-source network scanner designed for performing fast Internet-wide IPv4 and IPv6 network research scanning. XMap was initially developed as the research artifact of a paper published at 2021 IEEE/IFIP International Conference on Dependable Systems and Networks (DSN '21) and then made available on GitHub. XMap is the first tool to support fast Internet-wide IPv6 network scanning in 2020. During the last five years, XMap has made substantial impact in academia, industry, and government. It has been referenced in 52 research papers (15 published at top-tier security venues and 11 in leading networking societies), received over 450 GitHub stars, featured in multiple news outlets, and deployed or recommended by international companies up to date. Additionally, XMap has contributed to the implementation of RFC documents and the discovery of various vulnerabilities. This paper provides fundamental details about XMap, its architecture, and its impact.

cs.CR↗

arXiv2026

GhostCite: A Large-Scale Analysis of Citation Validity in the Age of Large Language Models

Citations provide the basis for trusting scientific claims; when they are invalid or fabricated, this trust collapses. With the advent of Large Language Models (LLMs), this risk has intensified: LLMs are increasingly used for academic writing, but their tendency to fabricate citations (``ghost citations'') poses a systemic threat to citation validity. To quantify this threat, we develop \citeb, an open-source framework for large-scale citation verification, and conduct a comprehensive study of citation validity in the LLM era through three complementary experiments. First, we benchmark 13 LLMs on citation generation task in various research domains, finding that all models hallucinate citations at rate from 14.23\% to 94.93\%. Second, we analyze 2.2 million citations from 56,381 papers at AI/ML and Security venues (2020--2025), finding that 1.07\% of papers contain invalid citations, with an 80.9\% increase in 2025. Third, we survey 97 researchers, finding that 87.2\% use AI-powered tools in their workflows, 76.7\% of reviewers do not thoroughly check references, and 74.5\% view peer review as ineffective at catching citation errors. Based on these findings, we argue that ghost citations represent a systemic threat to academic integrity, and call for coordinated efforts from community to address this challenge.

cs.CR↗

arXiv2026

Quaternion optical computing chip for parallel high-dimensional data processing

Optical computing chips have emerged as a transformative computing technology due to their high computational density, low energy consumption, and compact footprint. While real- and complex-valued computing chips have been well developed, their fundamental limitations in representing high-dimensional data significantly constrain their applicability in modern signal processing. Quaternions enable direct operations on three- and four-dimensional data, powering high-dimensional processing in data analytics and artificial intelligence. Here we demonstrate a quaternion optical computing chip (QOCC) for the first time and benchmark its performance in several typical application scenarios: three-dimensional point cloud processing, RGB chromatic transformation, and quaternion convolutional neural network for color image recognition. The QOCC harnesses high parallelism of light by wavelength-division multiplexing, processing high-dimensional data simultaneously through multiple optical wavelength channels. Compared to the electronic computing counterpart, our QOCC achieves higher computational fidelity (root mean square error < 0.035) and substantially reduced computational load (2/3 lower). It paves the way towards next-generation optical computing, overcoming the limitations of traditional computing systems in high-dimensional data processing.

physics.optics↗

arXiv2025

LLM Agent Meets Agentic AI: Can LLM Agents Simulate Customers to Evaluate Agentic-AI-based Shopping Assistants?

Agentic AI is emerging, capable of executing tasks through natural language, such as Copilot for coding or Amazon Rufus for shopping. Evaluating these systems is challenging, as their rapid evolution outpaces traditional human evaluation. Researchers have proposed LLM Agents to simulate participants as digital twins, but it remains unclear to what extent a digital twin can represent a specific customer in multi-turn interaction with an agentic AI system. In this paper, we recruited 40 human participants to shop with Amazon Rufus, collected their personas, interaction traces, and UX feedback, and then created digital twins to repeat the task. Pairwise comparison of human and digital-twin traces shows that while agents often explored more diverse choices, their action patterns aligned with humans and yielded similar design feedback. This study is the first to quantify how closely LLM agents can mirror human multi-turn interaction with an agentic AI system, highlighting their potential for scalable evaluation.

cs.HC↗

arXiv2025

Electrically Reconfigurable Arbitrary Splitting-Ratio Optical Splitter Based on Low-Loss Sb2Se3

Reconfigurable beam splitters capable of being arbitrarily programmed for the power splitting ratios are vital for the adaptive optical networks and photonic computing. Conventional mechanisms such as thermo-optic, free-carrier, or mechanical tuning are usually volatile and require continuous power, limiting their suitability for low-frequency and low power-consumption programmable operations. Here, we experimentally demonstrate an electrically reconfigurable beam splitter based on the low-loss phase-change material Sb2Se3, enabling multi-level and arbitrary splitting-ratio (SR) control. By locally triggering phase transitions in the coupling region with integrated micro-electrodes, we exploit the high refractive-index contrast between different phases and negligible absorption in the near-infrared wavelength of Sb2Se3 to precisely tune the coupling strength with non-volatile retention. 8-level of power splitting states is achieved within a compact footprint of ~14.5-{\mu}m in the experiments, with insertion loss is ~1 dB across 1515-1550 nm and near-zero static power. Combining the advantages of compactness, broad bandwidth, low loss, non-volatility, and multi-level control experimentally, this device provides a universal building block for scalable, energy-efficient reconfigurable photonic circuits, with great prospects in optical computing and intelligent communication systems.

physics.optics↗

arXiv2025

Wavelet-based Decoupling Framework for low-light Stereo Image Enhancement

Low-light images suffer from complex degradation, and existing enhancement methods often encode all degradation factors within a single latent space. This leads to highly entangled features and strong black-box characteristics, making the model prone to shortcut learning. To mitigate the above issues, this paper proposes a wavelet-based low-light stereo image enhancement method with feature space decoupling. Our insight comes from the following findings: (1) Wavelet transform enables the independent processing of low-frequency and high-frequency information. (2) Illumination adjustment can be achieved by adjusting the low-frequency component of a low-light image, extracted through multi-level wavelet decomposition. Thus, by using wavelet transform the feature space is decomposed into a low-frequency branch for illumination adjustment and multiple high-frequency branches for texture enhancement. Additionally, stereo low-light image enhancement can extract useful cues from another view to improve enhancement. To this end, we propose a novel high-frequency guided cross-view interaction module (HF-CIM) that operates within high-frequency branches rather than across the entire feature space, effectively extracting valuable image details from the other view. Furthermore, to enhance the high-frequency information, a detail and texture enhancement module (DTEM) is proposed based on cross-attention mechanism. The model is trained on a dataset consisting of images with uniform illumination and images with non-uniform illumination. Experimental results on both real and synthetic images indicate that our algorithm offers significant advantages in light adjustment while effectively recovering high-frequency information. The code and dataset are publicly available at: https://github.com/Cherisherr/WDCI-Net.git.

cs.CV↗

arXiv2025

Do Vision-Language Models Have Internal World Models? Towards an Atomic Evaluation

Internal world models (WMs) enable agents to understand the world's state and predict transitions, serving as the basis for advanced deliberative reasoning. Recent large Vision-Language Models (VLMs), such as OpenAI o3, GPT-4o and Gemini, exhibit potential as general-purpose WMs. While the latest studies have evaluated and shown limitations in specific capabilities such as visual understanding, a systematic evaluation of VLMs' fundamental WM abilities remains absent. Drawing on comparative psychology and cognitive science, we propose a two-stage framework that assesses Perception (visual, spatial, temporal, quantitative, and motion) and Prediction (mechanistic simulation, transitive inference, compositional inference) to provide an atomic evaluation of VLMs as WMs. Guided by this framework, we introduce WM-ABench, a large-scale benchmark comprising 23 fine-grained evaluation dimensions across 6 diverse simulated environments with controlled counterfactual simulations. Through 660 experiments on 15 latest commercial and open-source VLMs, we find that these models exhibit striking limitations in basic world modeling abilities. For instance, almost all models perform at near-random accuracy when distinguishing motion trajectories. Additionally, they lack disentangled understanding -- e.g., some models tend to believe blue objects move faster than green ones. More rich results and analyses reveal significant gaps between VLMs and human-level world modeling.

cs.CL↗

arXiv2025

Detection of Ultra-Trace Heavy metals in Aerosols with pg^m3 Sensitivity Using Filament-Induced Fluorescence Spectroscopy

Heavy metal pollution, particularly in the form of airborne aerosols such as lead (Pb), cadmium (Cd), mercury (Hg), and cobalt (Co), poses serious health and environmental risks, necessitating highly sensitive remote detection techniques. In this study, Filament-Induced Fluorescence Spectroscopy (FIFS) was employed to detect ultra-trace concentrations of heavy metal aerosols with high sensitivity and stability. By systematically optimizing the balance between filament length and detection distance, the optimal detection distance under the current experimental conditions was determined. With a detection distance of 10 m, this work achieved a minimum detectable concentration of 0.47 pg m^-3 for Pb and an extrapolated limit of detection (LOD) of 0.3 pg m^-3, with excellent signal stability (RSD < 7%) over a concentration range from 0.47 pg m^-3 to 0.47 g m^-3. Additionally, Cd, Hg, and Co aerosols were also successfully detected under the same conditions, with detection limits of 2 pg m^-3, 0.25 pg m^-3, and 3 pg m^-3, respectively, further confirming the versatility of FIFS in detecting diverse heavy metals. Theoretical predictions suggest that increasing laser power could further enhance the detection capability. These results highlight the ultra-sensitive remote detection capability of FIFS for heavy metal aerosol detection and provide valuable insights for optimizing system parameters to enhance its application performance in environmental monitoring.

physics.ins-det↗

arXiv2025

Intense THz s-SNOM for nonlinearity engineering in nanoscale

Terahertz (THz) nonlinear optics offer powerful tools to investigate and manipulate electronic dynamics in condensed matter. Confining high-peak-power THz pulses within near field can effectively generates extremely localized electromagnetic fields in spatio-temporal, enabling to precisely explore and control carrier transient dynamics from THz nonlinearity perspective. However, the combination of the high peak power THz pulses and the near-field optic techniques remains challenging due to the incompatibility between low repetition THz pulses and typical near-field demodulation schemes. Here, we construct high peak power THz scattering scanning near-field microscopy (THz s-SNOM) by combining THz pulses emitted from two-color femtosecond laser filaments with a tapping mode atomic force microscopy (AFM) and explore efficient THz third harmonics generation (THG) from the Cd3As2 film in nanoscale. The power-law dependence of the THz harmonics and theoretical calculation reveals a convincing third harmonic generation that is attributed to the nonequilibrium intraband dynamics driven by the strong THz pulses. Especially, the nanoscopic near-field THz third harmonic imaging with resolution of 200 nm ({\lambda}/3000) of 3D Dirac semimetal are demonstrated. The high peak power THz s-SNOM can provide a great platform for exploring and manipulating the nonlinear physics, carrier dynamics and quantum coherent phenomena driven by the localized THz field with nanoscale resolution, thereby guiding the development of the integrated high-performance nonlinear photonic devices.

physics.optics↗

arXiv2025

Arbitrary control of the flow of light using pseudomagnetic fields in photonic crystals at telecommunication wavelengths

In photonics, the idea of controlling light in a similar way that magnetic fields control electrons has always been attractive. It can be realized by synthesizing pseudomagnetic fields (PMFs) in photonic crystals (PhCs). Previous works mainly focus on the Landau levels and the robust transport of the chiral states. More versatile control over light using complex nonuniform PMFs such as the flexible splitting and routing of light has been elusive, which hinders their application in practical photonic integrated circuits. Here we propose an universal and systematic methodology to design nonuniform PMFs and arbitrarily control the flow of light in silicon PhCs at telecommunication wavelengths. As proofs of concept, a low-loss S-bend and a highly efficient 50:50 power splitter based on PMFs are experimentally demonstrated. A high-speed data transmission experiment is performed on these devices to prove their applicability in real communication systems. The proposed method offers a new paradigm for the exploration of fundamental physics and the development of novel nanophotonic devices.

physics.optics↗