arXiv subjects

Hang Zhang

Publications and source records attributed to Hang Zhang.

At least 91 records · Page 5Linked to original sources

arXiv2025

AR-Med: Automated Relevance Enhancement in Medical Search via LLM-Driven Information Augmentation

Accurate and reliable search on online healthcare platforms is critical for user safety and service efficacy. Traditional methods, however, often fail to comprehend complex and nuanced user queries, limiting their effectiveness. Large language models (LLMs) present a promising solution, offering powerful semantic understanding to bridge this gap. Despite their potential, deploying LLMs in this high-stakes domain is fraught with challenges, including factual hallucinations, specialized knowledge gaps, and high operational costs. To overcome these barriers, we introduce \textbf{AR-Med}, a novel framework for \textbf{A}utomated \textbf{R}elevance assessment for \textbf{Med}ical search that has been successfully deployed at scale on the Online Medical Delivery Platforms. AR-Med grounds LLM reasoning in verified medical knowledge through a retrieval-augmented approach, ensuring high accuracy and reliability. To enable efficient online service, we design a practical knowledge distillation scheme that compresses large teacher models into compact yet powerful student models. We also introduce LocalQSMed, a multi-expert annotated benchmark developed to guide model iteration and ensure strong alignment between offline and online performance. Extensive experiments show AR-Med achieves an offline accuracy of over 93\%, a 24\% absolute improvement over the original online system, and delivers significant gains in online relevance and user satisfaction. Our work presents a practical and scalable blueprint for developing trustworthy, LLM-powered systems in real-world healthcare applications.

cs.CL↗

arXiv2025

Qwen3-VL Technical Report

We introduce Qwen3-VL, the most capable vision-language model in the Qwen series to date, achieving superior performance across a broad range of multimodal benchmarks. It natively supports interleaved contexts of up to 256K tokens, seamlessly integrating text, images, and video. The model family includes both dense (2B/4B/8B/32B) and mixture-of-experts (30B-A3B/235B-A22B) variants to accommodate diverse latency-quality trade-offs. Qwen3-VL delivers three core pillars: (i) markedly stronger pure-text understanding, surpassing comparable text-only backbones in several cases; (ii) robust long-context comprehension with a native 256K-token window for both text and interleaved multimodal inputs, enabling faithful retention, retrieval, and cross-referencing across long documents and videos; and (iii) advanced multimodal reasoning across single-image, multi-image, and video tasks, demonstrating leading performance on comprehensive evaluations such as MMMU and visual-math benchmarks (e.g., MathVista and MathVision). Architecturally, we introduce three key upgrades: (i) an enhanced interleaved-MRoPE for stronger spatial-temporal modeling across images and video; (ii) DeepStack integration, which effectively leverages multi-level ViT features to tighten vision-language alignment; and (iii) text-based time alignment for video, evolving from T-RoPE to explicit textual timestamp alignment for more precise temporal grounding. Under comparable token budgets and latency constraints, Qwen3-VL achieves superior performance in both dense and Mixture-of-Experts (MoE) architectures. We envision Qwen3-VL serving as a foundational engine for image-grounded reasoning, agentic decision-making, and multimodal code intelligence in real-world workflows.

cs.CV↗

arXiv2025

Splats in Splats: Robust and Effective 3D Steganography towards Gaussian Splatting

3D Gaussian splatting (3DGS) has demonstrated impressive 3D reconstruction performance with explicit scene representations. Given the widespread application of 3DGS in 3D reconstruction and generation tasks, there is an urgent need to protect the copyright of 3DGS assets. However, existing copyright protection techniques for 3DGS overlook the usability of 3D assets, posing challenges for practical deployment. Here we describe splats in splats, the first 3DGS steganography framework that embeds 3D content in 3DGS itself without modifying any attributes. To achieve this, we take a deep insight into spherical harmonics (SH) and devise an importance-graded SH coefficient encryption strategy to embed the hidden SH coefficients. Furthermore, we employ a convolutional autoencoder to establish a mapping between the original Gaussian primitives' opacity and the hidden Gaussian primitives' opacity. Extensive experiments indicate that our method significantly outperforms existing 3D steganography techniques, with 5.31% higher scene fidelity and 3x faster rendering speed, while ensuring security, robustness, and user experience.

cs.CV↗

arXiv2025

Plug-and-Play Clarifier: A Zero-Shot Multimodal Framework for Egocentric Intent Disambiguation

The performance of egocentric AI agents is fundamentally limited by multimodal intent ambiguity. This challenge arises from a combination of underspecified language, imperfect visual data, and deictic gestures, which frequently leads to task failure. Existing monolithic Vision-Language Models (VLMs) struggle to resolve these multimodal ambiguous inputs, often failing silently or hallucinating responses. To address these ambiguities, we introduce the Plug-and-Play Clarifier, a zero-shot and modular framework that decomposes the problem into discrete, solvable sub-tasks. Specifically, our framework consists of three synergistic modules: (1) a text clarifier that uses dialogue-driven reasoning to interactively disambiguate linguistic intent, (2) a vision clarifier that delivers real-time guidance feedback, instructing users to adjust their positioning for improved capture quality, and (3) a cross-modal clarifier with grounding mechanism that robustly interprets 3D pointing gestures and identifies the specific objects users are pointing to. Extensive experiments demonstrate that our framework improves the intent clarification performance of small language models (4--8B) by approximately 30%, making them competitive with significantly larger counterparts. We also observe consistent gains when applying our framework to these larger models. Furthermore, our vision clarifier increases corrective guidance accuracy by over 20%, and our cross-modal clarifier improves semantic answer accuracy for referential grounding by 5%. Overall, our method provides a plug-and-play framework that effectively resolves multimodal ambiguity and significantly enhances user experience in egocentric interaction.

cs.HC↗

arXiv2025

UniMapGen: A Generative Framework for Large-Scale Map Construction from Multi-modal Data

Large-scale map construction plays a vital role in applications like autonomous driving and navigation systems. Traditional large-scale map construction approaches mainly rely on costly and inefficient special data collection vehicles and labor-intensive annotation processes. While existing satellite-based methods have demonstrated promising potential in enhancing the efficiency and coverage of map construction, they exhibit two major limitations: (1) inherent drawbacks of satellite data (e.g., occlusions, outdatedness) and (2) inefficient vectorization from perception-based methods, resulting in discontinuous and rough roads that require extensive post-processing. This paper presents a novel generative framework, UniMapGen, for large-scale map construction, offering three key innovations: (1) representing lane lines as \textbf{discrete sequence} and establishing an iterative strategy to generate more complete and smooth map vectors than traditional perception-based methods. (2) proposing a flexible architecture that supports \textbf{multi-modal} inputs, enabling dynamic selection among BEV, PV, and text prompt, to overcome the drawbacks of satellite data. (3) developing a \textbf{state update} strategy for global continuity and consistency of the constructed large-scale map. UniMapGen achieves state-of-the-art performance on the OpenSatMap dataset. Furthermore, UniMapGen can infer occluded roads and predict roads missing from dataset annotations. Our code will be released.

cs.CV↗

arXiv2025

LLMBisect: Breaking Barriers in Bug Bisection with A Comparative Analysis Pipeline

Bug bisection has been an important security task that aims to understand the range of software versions impacted by a bug, i.e., identifying the commit that introduced the bug. However, traditional patch-based bisection methods are faced with several significant barriers: For example, they assume that the bug-inducing commit (BIC) and the patch commit modify the same functions, which is not always true. They often rely solely on code changes, while the commit message frequently contains a wealth of vulnerability-related information. They are also based on simple heuristics (e.g., assuming the BIC initializes lines deleted in the patch) and lack any logical analysis of the vulnerability. In this paper, we make the observation that Large Language Models (LLMs) are well-positioned to break the barriers of existing solutions, e.g., comprehend both textual data and code in patches and commits. Unlike previous BIC identification approaches, which yield poor results, we propose a comprehensive multi-stage pipeline that leverages LLMs to: (1) fully utilize patch information, (2) compare multiple candidate commits in context, and (3) progressively narrow down the candidates through a series of down-selection steps. In our evaluation, we demonstrate that our approach achieves significantly better accuracy than the state-of-the-art solution by more than 38\%. Our results further confirm that the comprehensive multi-stage pipeline is essential, as it improves accuracy by 60\% over a baseline LLM-based bisection method.

cs.LG↗

arXiv2025

LineBreaker: Finding Token-Inconsistency Bugs with Large Language Models

Token-inconsistency bugs (TIBs) involve the misuse of syntactically valid yet incorrect code tokens, such as misused variables and erroneous function invocations, which can often lead to software bugs. Unlike simple syntactic bugs, TIBs occur at the semantic level and are subtle - sometimes they remain undetected for years. Traditional detection methods, such as static analysis and dynamic testing, often struggle with TIBs due to their versatile and context-dependent nature. However, advancements in large language models (LLMs) like GPT-4 present new opportunities for automating TIB detection by leveraging these models' semantic understanding capabilities. This paper reports the first systematic measurement of LLMs' capabilities in detecting TIBs, revealing that while GPT-4 shows promise, it exhibits limitations in precision and scalability. Specifically, its detection capability is undermined by the model's tendency to focus on the code snippets that do not contain TIBs; its scalability concern arises from GPT-4's high cost and the massive amount of code requiring inspection. To address these challenges, we introduce \name, a novel and cascaded TIB detection system. \name leverages smaller, code-specific, and highly efficient language models to filter out large numbers of code snippets unlikely to contain TIBs, thereby significantly enhancing the system's performance in terms of precision, recall, and scalability. We evaluated \name on 154 Python and C GitHub repositories, each with over 1,000 stars, uncovering 123 new flaws, 45\% of which could be exploited to disrupt program functionalities. Out of our 69 submitted fixes, 41 have already been confirmed or merged.

cs.CR↗

arXiv2025

High-pass filtered fidelity-imposed network edit (HP-FINE) for robust quantitative susceptibility mapping from high-pass filtered phase

Purpose: To improve the generalization ability of deep learning based predictions of quantitative susceptibility mapping (QSM) from high-pass filtered phase (HPFP) data. Methods: A network fine-tuning step called HP-FINE is proposed, which is based on the high-pass filtering forward model with low-frequency preservation regularization. Several comparisons were conducted: 1. HP-FINE with and without low-frequency regularization, 2. three 3D network architectures (Unet, Progressive Unet, and Big Unet), 3. two types of network output (recovered field and susceptibility), and 4. pre-training with and without the filtering augmentation. HPFP datasets with diverse high-pass filters, another acquisition voxel size, and prospective acquisition were used to assess the accuracy of QSM predictions. In the retrospective datasets, quantitative metrics (PSNR, SSIM, RMSE and HFEN) were used for evaluation. In the prospective dataset, statistics of ROI linear regression and Bland-Altman analysis were used for evaluation. Results: In the retrospective datasets, adding low-frequency regularization in HP-FINE substantially improved prediction accuracy compared to the pre-trained results, especially when combined with the filtering augmentation and recovered field output. In the prospective datasets, HP-FINE with low-frequency regularization and recovered field output demonstrated the preservation of ROI values, a result that was not achieved when using susceptibility as the output. Furthermore, Progressive Unet pre-trained with a combination of multiple losses outperformed both Unet and Progressive Unet pre-trained with a single loss in terms of preserving ROI values.

eess.IV↗

arXiv2025

RimSet: Quantitatively Identifying and Characterizing Chronic Active Multiple Sclerosis Lesion on Quantitative Susceptibility Maps

Background: Rim+ lesions in multiple sclerosis (MS), detectable via Quantitative Susceptibility Mapping (QSM), correlate with increased disability. Existing literature lacks quantitative analysis of these lesions. We introduce RimSet for quantitative identification and characterization of rim+ lesions on QSM. Methods: RimSet combines RimSeg, an unsupervised segmentation method using level-set methodology, and radiomic measurements with Local Binary Pattern texture descriptors. We validated RimSet using simulated QSM images and an in vivo dataset of 172 MS subjects with 177 rim+ and 3986 rim-lesions. Results: RimSeg achieved a 78.7% Dice score against the ground truth, with challenges in partial rim lesions. RimSet detected rim+ lesions with a partial ROC AUC of 0.808 and PR AUC of 0.737, surpassing existing methods. QSMRim-Net showed the lowest mean square error (0.85) and high correlation (0.91; 95% CI: 0.88, 0.93) with expert annotations at the subject level.

eess.IV↗

arXiv2025

MSRepaint: Multiple Sclerosis Repaint with Conditional Denoising Diffusion Implicit Model for Bidirectional Lesion Filling and Synthesis

In multiple sclerosis, lesions interfere with automated magnetic resonance imaging analyses such as brain parcellation and deformable registration, while lesion segmentation models are hindered by the limited availability of annotated training data. To address both issues, we propose MSRepaint, a unified diffusion-based generative model for bidirectional lesion filling and synthesis that restores anatomical continuity for downstream analyses and augments segmentation through realistic data generation. MSRepaint conditions on spatial lesion masks for voxel-level control, incorporates contrast dropout to handle missing inputs, integrates a repainting mechanism to preserve surrounding anatomy during lesion filling and synthesis, and employs a multi-view DDIM inversion and fusion pipeline for 3D consistency with fast inference. Extensive evaluations demonstrate the effectiveness of MSRepaint across multiple tasks. For lesion filling, we evaluate both the accuracy within the filled regions and the impact on downstream tasks including brain parcellation and deformable registration. MSRepaint outperforms the traditional lesion filling methods FSL and NiftySeg, and achieves accuracy on par with FastSurfer-LIT, a recent diffusion model-based inpainting method, while offering over 20 times faster inference. For lesion synthesis, state-of-the-art MS lesion segmentation models trained on MSRepaint-synthesized data outperform those trained on CarveMix-synthesized data or real ISBI challenge training data across multiple benchmarks, including the MICCAI 2016 and UMCL datasets. Additionally, we demonstrate that MSRepaint's unified bidirectional filling and synthesis capability, with full spatial control over lesion appearance, enables high-fidelity simulation of lesion evolution in longitudinal MS progression.

eess.IV↗

arXiv2025

ViewSpatial-Bench: Evaluating Multi-perspective Spatial Localization in Vision-Language Models

Vision-language models (VLMs) have demonstrated remarkable capabilities in understanding and reasoning about visual content, but significant challenges persist in tasks requiring cross-viewpoint understanding and spatial reasoning. We identify a critical limitation: current VLMs excel primarily at egocentric spatial reasoning (from the camera's perspective) but fail to generalize to allocentric viewpoints when required to adopt another entity's spatial frame of reference. We introduce ViewSpatial-Bench, the first comprehensive benchmark designed specifically for multi-viewpoint spatial localization recognition evaluation across five distinct task types, supported by an automated 3D annotation pipeline that generates precise directional labels. Comprehensive evaluation of diverse VLMs on ViewSpatial-Bench reveals a significant performance disparity: models demonstrate reasonable performance on camera-perspective tasks but exhibit reduced accuracy when reasoning from a human viewpoint. By fine-tuning VLMs on our multi-perspective spatial dataset, we achieve an overall performance improvement of 46.24% across tasks, highlighting the efficacy of our approach. Our work establishes a crucial benchmark for spatial intelligence in embodied AI systems and provides empirical evidence that modeling 3D spatial relationships enhances VLMs' corresponding spatial comprehension capabilities.

cs.CV↗

arXiv2025

MMR1: Enhancing Multimodal Reasoning with Variance-Aware Sampling and Open Resources

Large multimodal reasoning models have achieved rapid progress, but their advancement is constrained by two major limitations: the absence of open, large-scale, high-quality long chain-of-thought (CoT) data, and the instability of reinforcement learning (RL) algorithms in post-training. Group Relative Policy Optimization (GRPO), the standard framework for RL fine-tuning, is prone to gradient vanishing when reward variance is low, which weakens optimization signals and impairs convergence. This work makes three contributions: (1) We propose Variance-Aware Sampling (VAS), a data selection strategy guided by Variance Promotion Score (VPS) that combines outcome variance and trajectory diversity to promote reward variance and stabilize policy optimization. (2) We release large-scale, carefully curated resources containing ~1.6M long CoT cold-start data and ~15k RL QA pairs, designed to ensure quality, difficulty, and diversity, along with a fully reproducible end-to-end training codebase. (3) We open-source a family of multimodal reasoning models in multiple scales, establishing standardized baselines for the community. Experiments across mathematical reasoning benchmarks demonstrate the effectiveness of both the curated data and the proposed VAS. Comprehensive ablation studies and analyses provide further insight into the contributions of each component. In addition, we theoretically establish that reward variance lower-bounds the expected policy gradient magnitude, with VAS serving as a practical mechanism to realize this guarantee. Our code, data, and checkpoints are available at https://github.com/LengSicong/MMR1.

cs.CV↗

arXiv2025

Ideal Registration? Segmentation is All You Need

Deep learning has revolutionized image registration by its ability to handle diverse tasks while achieving significant speed advantages over conventional approaches. Current approaches, however, often employ globally uniform smoothness constraints that fail to accommodate the complex, regionally varying deformations characteristic of anatomical motion. To address this limitation, we propose SegReg, a Segmentation-driven Registration framework that implements anatomically adaptive regularization by exploiting region-specific deformation patterns. Our SegReg first decomposes input moving and fixed images into anatomically coherent subregions through segmentation. These localized domains are then processed by the same registration backbone to compute optimized partial deformation fields, which are subsequently integrated into a global deformation field. SegReg achieves near-perfect structural alignment (98.23% Dice on critical anatomies) using ground-truth segmentation, and outperforms existing methods by 2-12% across three clinical registration scenarios (cardiac, abdominal, and lung images) even with automatic segmentation. Our SegReg demonstrates a near-linear dependence of registration accuracy on segmentation quality, transforming the registration challenge into a segmentation problem. The source code will be released upon manuscript acceptance.

cs.CV↗

arXiv2025

SAMIR, an efficient registration framework via robust feature learning from SAM

Image registration is a fundamental task in medical image analysis. Deformations are often closely related to the morphological characteristics of tissues, making accurate feature extraction crucial. Recent weakly supervised methods improve registration by incorporating anatomical priors such as segmentation masks or landmarks, either as inputs or in the loss function. However, such weak labels are often not readily available, limiting their practical use. Motivated by the strong representation learning ability of visual foundation models, this paper introduces SAMIR, an efficient medical image registration framework that utilizes the Segment Anything Model (SAM) to enhance feature extraction. SAM is pretrained on large-scale natural image datasets and can learn robust, general-purpose visual representations. Rather than using raw input images, we design a task-specific adaptation pipeline using SAM's image encoder to extract structure-aware feature embeddings, enabling more accurate modeling of anatomical consistency and deformation patterns. We further design a lightweight 3D head to refine features within the embedding space, adapting to local deformations in medical images. Additionally, we introduce a Hierarchical Feature Consistency Loss to guide coarse-to-fine feature matching and improve anatomical alignment. Extensive experiments demonstrate that SAMIR significantly outperforms state-of-the-art methods on benchmark datasets for both intra-subject cardiac image registration and inter-subject abdomen CT image registration, achieving performance improvements of 2.68% on ACDC and 6.44% on the abdomen dataset. The source code will be publicly available on GitHub following the acceptance of this paper.

cs.CV↗

arXiv2025

Gaussian Primitive Optimized Deformable Retinal Image Registration

Deformable retinal image registration is notoriously difficult due to large homogeneous regions and sparse but critical vascular features, which cause limited gradient signals in standard learning-based frameworks. In this paper, we introduce Gaussian Primitive Optimization (GPO), a novel iterative framework that performs structured message passing to overcome these challenges. After an initial coarse alignment, we extract keypoints at salient anatomical structures (e.g., major vessels) to serve as a minimal set of descriptor-based control nodes (DCN). Each node is modelled as a Gaussian primitive with trainable position, displacement, and radius, thus adapting its spatial influence to local deformation scales. A K-Nearest Neighbors (KNN) Gaussian interpolation then blends and propagates displacement signals from these information-rich nodes to construct a globally coherent displacement field; focusing interpolation on the top (K) neighbors reduces computational overhead while preserving local detail. By strategically anchoring nodes in high-gradient regions, GPO ensures robust gradient flow, mitigating vanishing gradient signal in textureless areas. The framework is optimized end-to-end via a multi-term loss that enforces both keypoint consistency and intensity alignment. Experiments on the FIRE dataset show that GPO reduces the target registration error from 6.2\,px to ~2.4\,px and increases the AUC at 25\,px from 0.770 to 0.938, substantially outperforming existing methods. The source code can be accessed via https://github.com/xintian-99/GPOreg.

cs.CV↗

arXiv2025

Generative Foundation Model for Structured and Unstructured Electronic Health Records

Electronic health records (EHRs) are rich clinical data sources but complex repositories of patient data, spanning structured elements (demographics, vitals, lab results, codes), unstructured clinical notes and other modalities of data. Harnessing this heterogeneity is critical for improving patient outcomes. Recent advances in large language models (LLMs) have enabled foundation models that can learn from multiple data modalities and support clinical tasks. However, most current approaches simply serialize numeric EHR data into text, which risks losing temporal and quantitative detail. We introduce Generative Deep Patient (GDP), a multimodal foundation model that natively encodes structured EHR time-series via a CNN-Transformer encoder and fuses it with unstructured EHRs through cross-modal attention into a LLaMA-based decoder. GDP is trained in two stages: (1) generative pretraining, where it learns to produce clinical narratives from raw patient timelines while also performing masked feature prediction (MFP) and next time-step prediction (NTP) to capture temporal dynamics; and (2) multi-task fine-tuning for clinically meaningful predictions (e.g., heart failure, type 2 diabetes, 30-day readmission). In clinical prediction, GDP demonstrated superior performance on MIMIC-IV: heart failure AUROC = 0.923, type 2 diabetes AUROC = 0.817, and 30-day readmission AUROC = 0.627. For narrative generation, GDP achieved ROUGE-L = 0.135 and BERTScore-F1 = 0.545. In a blinded human evaluation, GDP-Instruct scored highest on faithfulness, fluency, and overall clinical utility, suggesting reduced hospital documentation workload without sacrificing accuracy. Our results demonstrate that a single multimodal foundation model can both predict clinically actionable events and generate high-quality clinical narratives. Furthermore, GDP's flexible architecture can be extended to additional modalities.

cs.AI↗

arXiv2025

Frequency Point Game Environment for UAVs via Expert Knowledge and Large Language Model

Unmanned Aerial Vehicles (UAVs) have made significant advancements in communication stability and security through techniques such as frequency hopping, signal spreading, and adaptive interference suppression. However, challenges remain in modeling spectrum competition, integrating expert knowledge, and predicting opponent behavior. To address these issues, we propose UAV-FPG (Unmanned Aerial Vehicle - Frequency Point Game), a game-theoretic environment model that simulates the dynamic interaction between interference and anti-interference strategies of opponent and ally UAVs in communication frequency bands. The model incorporates a prior expert knowledge base to optimize frequency selection and employs large language models for path planning, simulating a "strong adversary". Experimental results highlight the effectiveness of integrating the expert knowledge base and the large language model, with the latter significantly improving path planning in dynamic scenarios through iterative interactions, outperforming fixed-path strategies. UAV-FPG provides a robust platform for advancing anti-jamming strategies and intelligent decision-making in UAV communication systems.

cs.MA↗

arXiv2025

Agent-Based Anti-Jamming Techniques for UAV Communications in Adversarial Environments: A Comprehensive Survey

Unmanned Aerial Vehicle communications are encountering increasingly severe multi-source interference challenges in dynamic adversarial environments, which impose higher demands on their reliability and resilience. To address these challenges, agent-based autonomous anti-jamming techniques have emerged as a crucial research direction. This paper presents a comprehensive survey that first formalizes the concept of intelligent anti-jamming agents for UAV communications and establishes a closed-loop decision-making framework centered on the "Perception-Decision-Action" (P-D-A) paradigm. Within this framework, we systematically review key technologies at each stage, with particular emphasis on employing game theory to model UAV-jammer interactions and integrating reinforcement learning-based intelligent algorithms to derive adaptive anti-jamming strategies. Furthermore, we discuss potential limitations of current approaches, identify critical engineering challenges, and outline promising future research directions, aiming to provide valuable references for developing more intelligent and robust anti-jamming communication systems for UAVs.

eess.SP↗