arXiv subjects

Yijun Wang

Publications and source records attributed to Yijun Wang.

At least 37 records · Page 2Linked to original sources

arXiv2026

Chat2Workflow: A Benchmark for Generating Executable Visual Workflows with Natural Language

At present, executable visual workflows have emerged as a mainstream paradigm in real-world industrial deployments, offering strong reliability and controllability. However, in current practice, such workflows are almost entirely constructed through manual engineering: developers must carefully design workflows, write prompts for each step, and repeatedly revise the logic as requirements evolve -- making development costly, time-consuming, and error-prone. To study whether large language models can automate this multi-round interaction process, we introduce Chat2Workflow, a benchmark for generating executable visual workflows directly from natural language, and propose a robust agentic baseline to improve performance. The benchmark is built from a large collection of real-world business workflows, with each instance designed so that the generated workflow can be transformed and directly deployed to practical workflow platforms such as Dify and Coze. Experimental results show that while state-of-the-art language models can often capture high-level intent, they struggle to generate correct, stable, and executable workflows, especially given complex and evolving requirements. Although our agentic baseline yields up to 6.05% resolve rate gains, the remaining real-world gap positions Chat2Workflow as a foundation for advancing industrial-grade automation. Code is available at https://github.com/zjunlp/Chat2Workflow.

cs.CL↗

arXiv2026

Towards Fine-Grained Robustness: Attention-Guided Test-Time Prompt Tuning for Vision-Language Models

Vision-Language Models (VLMs), such as CLIP, have achieved significant zero-shot performance on downstream tasks with various fine-tuning adaptation methods. However, recent studies have proven that adversarial attacks can significantly degrade the inference ability of VLMs, posing substantial risks to their practical applications. Prevalent test-time adaptation methods typically rely on multi-view augmentation to implement various fine-tuning strategies, which struggle to identify semantic information and are prone to destroying discriminative regions in fine-grained scenarios. To address these limitations, we propose Attention-Guided Test-Time Prompt Tuning (A-TPT), a semantics-preserving method designed for test-time adaptation. We first refine the gradient attention rollout mechanism to identify semantically meaningful regions surviving under adversarial attacks. Furthermore, we leverage them to guide the spatially varying augmentation intensities and multi-view ensemble for prompt tuning and inference. Extensive experiments demonstrate that A-TPT outperforms existing test-time adaptation methods on both adversarial and clean data. Codes are available at https://github.com/SEU-VIPGroup/A-TPT .

cs.CV↗

arXiv2026

Lightweight Cybersickness Detection based on User-Specific Eye and Head Tracking Data in Virtual Reality

The occurrence of cybersickness in virtual reality (VR) significantly impairs users' perception and sense of immersion. Therefore, timely detection of cybersickness and the application of appropriate intervention strategies are crucial for enhancing the user experience. However, existing cybersickness detection methods often suffer from issues such as poor detection reliability across different levels of cybersickness and unnecessary model complexity. Furthermore, while cybersickness exhibits significant inter-user variability, most existing approaches aggregate all data from users and lack user-specific solutions. In this paper, we investigate a lightweight approach for cybersickness detection incorporating an ensemble learning model and user-specific eye and head tracking data. Our experiments using the open-source dataset Simulation 2021 demonstrate that feature engineering and training set construction are critical for determining detection performance. Models trained with data from similar-content segments achieve the best results, attaining detection accuracies of 93% in the cross-user setting and 88% in the user-personalized setting, using only 23-dimensional eye and head features. Moreover, by using user-specific data, well-tuned ensemble learning models with shorter training and inference times can be feasibly applied to real-world cybersickness detection, offering superior time efficiency and outstanding detection performance. This work offers useful evidence toward the development of lightweight and user-adaptive cybersickness detection models for VR applications.

cs.HC↗

arXiv2026

GTA-2: Benchmarking General Tool Agents from Atomic Tool-Use to Open-Ended Workflows

The development of general-purpose agents requires a shift from executing simple instructions to completing complex, real-world productivity workflows. However, current tool-use benchmarks remain misaligned with real-world requirements, relying on AI-generated queries, dummy tools, and limited system-level coordination. To address this, we propose GTA-2, a hierarchical benchmark for General Tool Agents (GTA) spanning atomic tool use and open-ended workflows. Built on real-world authenticity, it leverages real user queries, deployed tools, and multimodal contexts. (i) GTA-Atomic, inherited from our prior GTA benchmark, evaluates short-horizon, closed-ended tool-use precision. (ii) GTA-Workflow introduces long-horizon, open-ended tasks for realistic end-to-end completion. To evaluate open-ended deliverables, we propose a recursive checkpoint-based evaluation mechanism that decomposes objectives into verifiable sub-goals, enabling unified evaluation of both model capabilities and agent execution frameworks (i.e., execution harnesses). Experiments reveal a pronounced capability cliff: while frontier models already struggle on atomic tasks (below 50%), they largely fail on workflows, with top models achieving only 14.39% success. Further analysis shows that checkpoint-guided feedback improves performance, while advanced frameworks such as Manus and OpenClaw substantially enhance workflow completion, highlighting the importance of execution harness design beyond the underlying model capacity. These findings provide guidance for developing reliable personal and professional assistants. Dataset and code will be available at https://github.com/open-compass/GTA.

cs.CL↗

arXiv2026

RAAP: Retrieval-Augmented Affordance Prediction with Cross-Image Action Alignment

Understanding object affordances is essential for enabling robots to perform purposeful and fine-grained interactions in diverse and unstructured environments. However, existing approaches either rely on retrieval, which is fragile due to sparsity and coverage gaps, or on large-scale models, which frequently mislocalize contact points and mispredict post-contact actions when applied to unseen categories, thereby hindering robust generalization. We introduce Retrieval-Augmented Affordance Prediction (RAAP), a framework that unifies affordance retrieval with alignment-based learning. By decoupling static contact localization and dynamic action direction, RAAP transfers contact points via dense correspondence and predicts action directions through a retrieval-augmented alignment model that consolidates multiple references with dual-weighted attention. Trained on compact subsets of DROID and HOI4D with as few as tens of samples per task, RAAP achieves consistent performance across unseen objects and categories, and enables zero-shot robotic manipulation in both simulation and the real world. Project website: https://github.com/SEU-VIPGroup/RAAP.

cs.RO↗

arXiv2026

AI assisted optimization of integrated waveguide polarizers containing 2D reduced graphene oxide

Reduced graphene oxide (rGO) exhibits strong anisotropic light absorption and high compatibility with photonic integrated chips, making it a promising material for implementing high performance onchip polarization selective devices. The performance of rGO integrated waveguide polarizers is highly dependent on the waveguide geometry, and achieving optimal performance requires exploring a large parameter space, making conventional mode simulation methods computationally demanding. Here, we propose and demonstrate a machine learning framework based on fully connected neural networks (FCNNs) to map the dependence of the polarizer figure of merit (FOM) on the waveguide geometry. Once trained by using a small dataset of low resolution mode simulation results, the FCNN framework can rapidly and accurately predict FOM values across a large structural parameter space with high resolution. Results show that this method can reduce overall computing time by more than 4 orders of magnitude as compared to the mode simulation methods, and achieve high prediction accuracy with an average deviation (AD) below 0.05. These results highlight the FCNN based machine learning framework as an efficient tool for the design and optimization of rGO integrated waveguide polarizers.

physics.optics↗

arXiv2026

MedEyes: Learning Dynamic Visual Focus for Medical Progressive Diagnosis

Accurate medical diagnosis often involves progressive visual focusing and iterative reasoning, characteristics commonly observed in clinical workflows. While recent vision-language models demonstrate promising chain-of-thought (CoT) reasoning capabilities via reinforcement learning with verifiable rewards (RLVR), their purely on-policy learning paradigm tends to reinforce superficially coherent but clinically inaccurate reasoning paths. We propose MedEyes, a novel reinforcement learning framework that dynamically models clinician-style diagnostic reasoning by progressively attending to and interpreting relevant medical image regions. By incorporating off-policy expert guidance, MedEyes converts expert visual search trajectories into structured external behavioral signals, guiding the model toward clinically aligned visual reasoning. We design the Gaze-guided Reasoning Navigator (GRN) to emulate the diagnostic process through a dual-mode exploration strategy, scanning for systematic abnormality localization and drilling for detailed regional analysis. To balance expert imitation and autonomous discovery, we introduce the Confidence Value Sampler (CVS), which employs nucleus sampling and adaptive termination to create diverse yet credible exploration paths. Finally, the dual-stream GRPO optimization framework decouples on-policy and off-policy learning signals, mitigating reward assimilation and entropy collapse. Experiments demonstrate that MedEyes achieves an average performance improvement of +8.5pp across multiple medical VQA benchmarks, validating MedEyes's potential in building trustworthy medical AI systems. Code is available at https://github.com/zhcz328/MedEyes.

cs.CV↗

arXiv2026

ChartE$^{3}$: A Comprehensive Benchmark for End-to-End Chart Editing

Charts are a fundamental visualization format for structured data analysis. Enabling end-to-end chart editing according to user intent is of great practical value, yet remains challenging due to the need for both fine-grained control and global structural consistency. Most existing approaches adopt pipeline-based designs, where natural language or code serves as an intermediate representation, limiting their ability to faithfully execute complex edits. We introduce ChartE$^{3}$, an End-to-End Chart Editing benchmark that directly evaluates models without relying on intermediate natural language programs or code-level supervision. ChartE$^{3}$ focuses on two complementary editing dimensions: local editing, which involves fine-grained appearance changes such as font or color adjustments, and global editing, which requires holistic, data-centric transformations including data filtering and trend line addition. ChartE$^{3}$ contains over 1,200 high-quality samples constructed via a well-designed data pipeline with human curation. Each sample is provided as a triplet of a chart image, its underlying code, and a multimodal editing instruction, enabling evaluation from both objective and subjective perspectives. Extensive benchmarking of state-of-the-art multimodal large language models reveals substantial performance gaps, particularly on global editing tasks, highlighting critical limitations in current end-to-end chart editing capabilities.

cs.CV↗

arXiv2026

RouteMoA: Dynamic Routing without Pre-Inference Boosts Efficient Mixture-of-Agents

Mixture-of-Agents (MoA) improves LLM performance through layered collaboration, but its dense topology raises costs and latency. Existing methods employ LLM judges to filter responses, yet still require all models to perform inference before judging, failing to cut costs effectively. They also lack model selection criteria and struggle with large model pools, where full inference is costly and can exceed context limits. To address this, we propose RouteMoA, an efficient mixture-of-agents framework with dynamic routing. It employs a lightweight scorer to perform initial screening by predicting coarse-grained performance from the query, narrowing candidates to a high-potential subset without inference. A mixture of judges then refines these scores through lightweight self- and cross-assessment based on existing model outputs, providing posterior correction without additional inference. Finally, a model ranking mechanism selects models by balancing performance, cost, and latency. RouteMoA outperforms MoA across varying tasks and model pool sizes, reducing cost by 89.8% and latency by 63.6% in the large-scale model pool.

cs.AI↗

arXiv2026

Quantifying the Upper Limit of Backflash Attack in Quantum Key Distribution

Quantum key distribution (QKD) provides information-theoretic security grounded in the fundamental laws of physics. Nevertheless, practical imperfections can introduce side channels that expose QKD systems to quantum hacking, especially passive attacks that are inherently difficult to detect. In this study, we experimentally and theoretically investigate the upper limit of the backflash attack-a representative passive side-channel threat. Using a fully equipped fiber-based QKD receiver, we demonstrate the feasibility of the attack and reveal its limited capability in distinguishing quantum states. We further develop a theoretical framework to quantify the maximum distinguishability achievable by an eavesdropper, taking into account the broadband spectral nature of backflash photons. The analysis shows that Eve can extract effective key information from at most 95.7% of the backflash photons. Based on these findings, we evaluate the secure key rate of a decoy-state BB84 QKD system under backflash attack. Our results provide a quantitative assessment of the vulnerability of QKD systems to backflash emissions and offer a general methodology to evaluate the practical security of QKD systems.

quant-ph↗

arXiv2025

fMRI2GES: Co-speech Gesture Reconstruction from fMRI Signal with Dual Brain Decoding Alignment

Understanding how the brain responds to external stimuli and decoding this process has been a significant challenge in neuroscience. While previous studies typically concentrated on brain-to-image and brain-to-language reconstruction, our work strives to reconstruct gestures associated with speech stimuli perceived by brain. Unfortunately, the lack of paired \{brain, speech, gesture\} data hinders the deployment of deep learning models for this purpose. In this paper, we introduce a novel approach, \textbf{fMRI2GES}, that allows training of fMRI-to-gesture reconstruction networks on unpaired data using \textbf{Dual Brain Decoding Alignment}. This method relies on two key components: (i) observed texts that elicit brain responses, and (ii) textual descriptions associated with the gestures. Then, instead of training models in a completely supervised manner to find a mapping relationship among the three modalities, we harness an fMRI-to-text model, a text-to-gesture model with paired data and an fMRI-to-gesture model with unpaired data, establishing dual fMRI-to-gesture reconstruction patterns. Afterward, we explicitly align two outputs and train our model in a self-supervision way. We show that our proposed method can reconstruct expressive gestures directly from fMRI recordings. We also investigate fMRI signals from different ROIs in the cortex and how they affect generation results. Overall, we provide new insights into decoding co-speech gestures, thereby advancing our understanding of neuroscience and cognitive science.

cs.AI↗

arXiv2025

Pathology-Aware Prototype Evolution via LLM-Driven Semantic Disambiguation for Multicenter Diabetic Retinopathy Diagnosis

Diabetic retinopathy (DR) grading plays a critical role in early clinical intervention and vision preservation. Recent explorations predominantly focus on visual lesion feature extraction through data processing and domain decoupling strategies. However, they generally overlook domain-invariant pathological patterns and underutilize the rich contextual knowledge of foundation models, relying solely on visual information, which is insufficient for distinguishing subtle pathological variations. Therefore, we propose integrating fine-grained pathological descriptions to complement prototypes with additional context, thereby resolving ambiguities in borderline cases. Specifically, we propose a Hierarchical Anchor Prototype Modulation (HAPM) framework to facilitate DR grading. First, we introduce a variance spectrum-driven anchor prototype library that preserves domain-invariant pathological patterns. We further employ a hierarchical differential prompt gating mechanism, dynamically selecting discriminative semantic prompts from both LVLM and LLM sources to address semantic confusion between adjacent DR grades. Finally, we utilize a two-stage prototype modulation strategy that progressively integrates clinical knowledge into visual prototypes through a Pathological Semantic Injector (PSI) and a Discriminative Prototype Enhancer (DPE). Extensive experiments across eight public datasets demonstrate that our approach achieves pathology-guided prototype evolution while outperforming state-of-the-art methods. The code is available at https://github.com/zhcz328/HAPM.

cs.AI↗

arXiv2025

The Faintest, Extremely Variable X-ray Tidal Disruption Event from a Supermassive Black Hole Binary?

Tidal disruption events (TDEs), which occur when stars enter the tidal radii of supermassive black holes (SMBHs) and are subsequently torn apart by their tidal forces, represent intriguing phenomena that stimulate growing research interest and pose an increasing number of puzzles in the era of time-domain astronomy. Here we report an unusual X-ray transient, XID 935, discovered in the 7 Ms Chandra Deep Field-South, the deepest X-ray survey ever. XID 935 experienced an overall X-ray dimming by a factor of more than 40 between 1999 and 2016. Not monotonically decreasing during this period, its X-ray luminosity increased by a factor $> 27$ within 2 months, from $L_{\rm 0.5-7\ keV}<10^{40.87}$ erg s$^{-1}$ (10 October 2014 -- 4 January 2015) to $L_{\rm 0.5-7\ keV}=10^{42.31\pm 0.20}$ erg s$^{-1}$ (16 March 2015). The X-ray position of XID 935 is located at the center of its host galaxy with a spectroscopic redshift of 0.251, whose optical spectra do not display emission characteristics associated with an active galactic nucleus. The peak 0.5--2.0 keV flux is the faintest among all the X-ray-selected TDE candidates to date. Thanks to a total exposure of $\sim 9.5$ Ms in the X-ray bands, we manage to secure relatively well-sampled, 20-year-long X-ray light curves of this deepest X-ray-selected TDE candidate. We find that a partial TDE model could not explain the main declining trend. An SMBH binary TDE model is in acceptable accordance with the light curves of XID 935; however, it fails to match short-timescale fluctuations exactly. Therefore, the exceptional observational features of XID 935 provide a key benchmark for refining quantitative TDE models and simulations.

astro-ph.HE↗

arXiv2025

Fast inflowing ionized absorber tracing the gas dynamics at sub-parsec scale around Mrk 3

Accretion onto supermassive black hole (SMBH) can release energy via radiation, jets or winds, providing feedback effects on the circumnuclear gas environment. However, not all active galactic nuclei (AGNs) exhibit clear signature of such feedback, and the dynamics of accreting gas on the inner sub-parsec scales remains poorly understood. Using high-resolution Chandra X-ray grating spectra of Mrk 3, we detect a fast inflowing ionized absorber characterized by redshifted Fe XXV and Fe XXVI absorption lines with confidence level in the $94-99.6\%$ range. Photoionization modeling reveals the inflowing absorber is located at $\lesssim0.04-0.74\rm~pc$, with red-shifted velocity decreasing from $6.1\pm0.5\times10^3\rm~km~s^{-1}$ to $3.4\pm0.3\times10^3\rm~km~s^{-1}$ over 11 years. Only $\sim0.6$\%--$3$\% of the inflowing material is estimated to reach the event horizon. This direct evidence of sub-parsec scale fueling inflow bridges the gap between the torus and the outer accretion disk. Additionally, a $0.86$-keV gas component with sub-solar metallicity ($Z\sim0.22$), outflowing at a velocity of $\sim330\rm~km~s^{-1}$, is detected in the soft X-ray band from XMM-Newton Reflection Grating Spectrometer, probably corresponding to shocked interstellar medium in the narrow-line region (NLR). The simultaneous presence of the apparent decelerating sub-parsec inflow and the NLR outflow favors a coherent scenario where a putative disk wind or broad-line region clouds may impede or even eject the accretion material, although other possibilities cannot be fully excluded.

astro-ph.HE↗

arXiv2025

Lithium Faraday Filter: Some Like It Hot

Magnetically induced rotation of linearly polarized light near an atomic resonance, combined with Doppler-broadened absorption windows, enables narrowband transmission of optical frequencies. An ultra-narrowband lithium vapor Faraday filter at about 671 nm is investigated experimentally and theoretically. The resulting Faraday filter transmittance is demonstrated using a lithium heat pipe oven under longitudinal magnetic fields ranging from 0 to 300 G. Optimization of the lithium Faraday filter performance reveals an optimal operating point at 264 °C and an external magnetic field of 269 G, yielding a peak transmission of approx. 82%. The lithium D$_1$- and D$_2$-transitions are only 10 GHz apart and temperature broadening leads to an overlap of the isotopes D-lines. Thus, the applied theoretical model needs to consider both transitions simultaneously. For this purpose, we extended an existing Python library (ElecSus), which now allows for the calculation of the atomic susceptibilities of lithium.

physics.atom-ph↗

arXiv2025

Warm absorber outflows in radio-loud active galactic nucleus 3C~59

Both jets and ionized outflows in active galactic nuclei (AGNs) are thought to play important roles in affecting the star formation and evolution of host galaxies, but their relationship is still unclear. As a pilot study, we performed a detailed spectral analysis for a radio-loud (RL) AGN 3C~59 ($z=0.1096$) by systematically considering various factors that may affect the fitting results, and thereby establishing a general spectral fitting strategy for subsequent research with larger sample. 3C~59 is one rare target for simultaneously studying jets and warm absorbers (WAs) that is one type of ionized outflows. Based on the multi-wavelength data from near-infrared (NIR) to hard X-ray bands detected by DESI, GALEX, and XMM-Newton, we used SPEX code to build broadband continuum models and perform photoionization modeling with PION code to constrain the physical parameters of WAs in 3C~59. We found two WAs with ionization parameter of $\log [ξ/(\rm{erg\ cm\ s}^{-1})] = 2.65^{+0.10}_{-0.09}$ and $1.65\pm 0.11$, respectively, and their outflowing velocities are $v_{\rm out} = -528^{+163}_{-222}\ \rm{km\ s}^{-1}$ and $-228^{+121}_{-122}\ \rm{km\ s}^{-1}$, respectively. These WAs are located between outer torus and narrow (emission-)line region, and their positive $v_{\rm out}$-$ξ$ relation can be explained by the radiation-pressure-driven mechanism. We found that the estimations of these physical properties are affected by the different spectral fitting strategies, such as the inclusion of NIR to ultra-violet data, the choice of energy range of spectrum, or the composition of the spectral energy distribution. Based on the same fitting strategy, this work presents a comparative study of outflow driven mechanism between a RL AGN (3C 59) and a radio-quiet AGN (NGC 3227), which suggests a similar driven mechanism of their WA outflows and a negligible role of jets in this process.

astro-ph.GA↗

arXiv2025

On the Perception Bottleneck of VLMs for Chart Understanding

Chart understanding requires models to effectively analyze and reason about numerical data, textual elements, and complex visual components. Our observations reveal that the perception capabilities of existing large vision-language models (LVLMs) constitute a critical bottleneck in this process. In this study, we delve into this perception bottleneck by decomposing it into two components: the vision encoder bottleneck, where the visual representation may fail to encapsulate the correct information, and the extraction bottleneck, where the language model struggles to extract the necessary information from the provided visual representations. Through comprehensive experiments, we find that (1) the information embedded within visual representations is substantially richer than what is typically captured by linear extractors, such as the widely used retrieval accuracy metric; (2) While instruction tuning effectively enhances the extraction capability of LVLMs, the vision encoder remains a critical bottleneck, demanding focused attention and improvement. Therefore, we further enhance the visual encoder to mitigate the vision encoder bottleneck under a contrastive learning framework. Empirical results demonstrate that our approach significantly mitigates the perception bottleneck and improves the ability of LVLMs to comprehend charts. Code is publicly available at https://github.com/hkust-nlp/Vision4Chart.

cs.CV↗

arXiv2025

Silicon Integrated Photonic Waveguide Polarizers with 2D MoS2 Films

Polarization control is of fundamental importance for modern optical systems, and optical polarizers serve as critical components for enabling this functionality. Here, we experimentally demonstrate optical polarizers by integrating 2D molybdenum disulfide (MoS2) films onto silicon photonic waveguides. High-quality monolayer MoS2 films with highly anisotropic light absorption are synthesized via a low-pressure chemical vapor deposition (LPCVD) method and subsequently transferred onto silicon-on-insulator (SOI) nanowire waveguides to fabricate integrated optical polarizers. Detailed measurements are carried out for the fabricated devices with various MoS2 film coating lengths and silicon waveguide geometry. The results show that a maximum polarization-dependent loss of ~21 dB is achieved, together with a high figure of merit of ~4.2. In addition, the hybrid waveguide polarizers exhibit broad operation bandwidth exceeding ~100 nm and excellent power durability. These results highlight the strong potential for on-chip integration of 2D MoS2 films to implement high-performance polarization selective devices.

physics.optics↗