arXiv subjects

Qiang Hu

Publications and source records attributed to Qiang Hu.

At least 19 recordsLinked to original sources

arXiv2026

CateKV: On Sequential Consistency for Long-Context LLM Inference Acceleration

Large language models (LLMs) have demonstrated strong capabilities in handling long-context tasks, but processing such long contexts remains challenging due to the substantial memory requirements and inference latency. In this work, we discover that certain attention heads exhibit sequential consistency in their attention patterns, which can be persistently identified using a coefficient-of-variation-based algorithm. Inspired by this observation, we propose CateKV, a hybrid KV cache method that retains only critical token information for consistent heads, thereby reducing KV cache size and computational overhead, while preserving the majority of KV pairs in adaptive heads to ensure high accuracy. We show the unique characteristics of our algorithm and its extension with existing acceleration methods. Comprehensive evaluations on long-context benchmarks show that, while maintaining accuracy comparable to full attention, CateKV reduces memory usage by up to $2.72\times$ and accelerates decoding by $2.18\times$ in single-sample inputs, and boosts throughput by $3.96\times$ in batch scenarios.

cs.LG↗

arXiv2026

The Setting of IMU Parameters in Kalman Filtering-based Information Fusion

The setting or tuning of specifications for the inertial measurement unit (IMU) is tricky in sensor fusion. The underneath conundrum is caused by the fact that the working condition of IMU is more complex than the stationary calibration scenario. Since the noises and biases instabilities calibrated under static condition cannot accommodate other cases, the effective tuning of IMU parameters largely hinges on the experience or profound understanding of the system. In the current work, the setting method of IMU parameters based on Allan variance calibration is delved into within the Kalman filtering framework. Specifically, the relationship between the power sepctral density and Allan variance is leveraged in formulating the process uncertainty in continuous-time filtering. Three typical IMU-based sensor fusion systems, including INS/GNSS integration, LiDAR-inertial odometry, and visual-inertial odometry are considered to show the feasibility and effectiveness of this parameter setting process.

cs.RO↗

arXiv2026

Linguistically-Aligned and Visually-Grounded Preference Optimization for Clinically-Augmented Medical Report Generation

Despite significant advances in Medical Report Generation (MRG), the reliability remains constrained by the prevalence of factual errors. While Direct Preference Optimization (DPO) has emerged as a promising post-training paradigm to enhance the performance of Supervised Fine-Tuned (SFT) MRG models, existing DPO-based MRG methods typically adopt a naive preference construction that directly pairs model-generated reports with ground truth reports. This strategy inadvertently entangles critical clinical findings with clinically irrelevant linguistic characteristics, and fundamentally lacks explicit vision-language alignment. To address these challenges, we propose DPO-Clin, a novel post-training framework that focuses preference optimization on clinical findings and cross-modal alignment. First, we introduce the Entity-level Clinical Diagnostic (ECD) module to perform a precise entity-level factual diagnosis. ECD guides the generation of linguistically-aligned report preference pairs, isolating clinical discrepancies from linguistic variations. Second, to achieve fine-grained cross-modal alignment, we develop M2DPO, a retrieval-augmented multi-modal DPO variant that enforces textual preference inversion triggered by visual context switches. Third, we locate correct yet highly uncertain predicted entities and apply counterfactual modifications to construct targeted preference data for latent risk mitigation, thereby further enhancing the model reliability. Extensive experiments on two public chest X-ray datasets (MIMIC-CXR and IU X-Ray) and an in-house endoscopy dataset demonstrate that DPO-Clin significantly improves the SFT baselines on clinical-aware metrics. Furthermore, it achieves superior performance over existing DPO-based MRG methods, exhibiting robust generalizability across distinct baseline architectures and diverse medical imaging modalities.

cs.CV↗

arXiv2026

CARVE: Cross-Slice Anisotropic Reallocation of Visual Evidence for Efficient 3D Medical Volume Understanding

Slice-based MLLMs leverage mature 2D encoders by representing 3D volumes as sequences of 2D slices. However, this slice-wise formulation produces thousands of visual tokens that burden the LLM backbone, many of which capture overlapping visual evidence across adjacent slices. To understand how effectively a growing visual token budget improves performance, we perform scaling analyses on two 3D medical VQA benchmarks and find diminishing returns: cost keeps rising while accuracy saturates, and improving in-plane resolution is more effective than adding slices at comparable budgets. The budget should therefore be allocated more selectively rather than simply enlarged, yet most token compression methods are designed for 2D images or videos, where redundancy arises from spatial layout or temporal motion rather than from near-duplicate content along the depth axis. We present CARVE, a training-free framework that compresses visual tokens prior to LLM inference and casts token reduction as budget-constrained 2.5D allocation. CARVE partitions the depth axis into coherent windows and allocates tokens non-uniformly according to normalized cross-slice evidence. Under a shared budget, CARVE builds spatial anchors on representative slices and retrieves locally varying evidence from the full volume, then merges remaining eligible tokens into nearby anchors within each window. Removing roughly 80% of the visual tokens on Hulu-Med-7B, CARVE leads all compression baselines on every AMOS-MM report-generation metric, with 6.2 points higher retention of full-token quality than the strongest baseline, and preserves 98.1% of full-token performance across three VQA benchmarks.

cs.CV↗

arXiv2026

MIEScore: Human-Aligned Evaluation for Multi-Source Image Editing

Recent advances in unified multimodal models have significantly improved text-guided image editing abilities. In particular, models such as Nano-Banana-Pro and GPT-Image-2 demonstrate emerging capabilities in multi-source image editing (MIE), including tasks such as object synthesis, person-background composition, and cross-image style fusion. However, existing benchmarks and image editing assessment (IEQA) methods remain primarily focused on single-image editing tasks and largely overlook the more challenging setting of MIE. This highlights the urgent need for a comprehensive and human-aligned benchmark for MIE. To this end, we introduce MIE-Bench, the first large-scale multiple image editing benchmark with fine-grained human preference annotations. Specifically, MIE-Bench includes 3,000 editing instances across 16 tasks, each involving more than two source images and an editing prompt, together with 36K edited images produced by 12 state-of-the-art editing models and over 108K mean opinion scores (MOSs) covering visual quality, instruction following, and attribute preservation. Based on MIE-Bench, we propose MIEScore, a multimodal large language model (MLLM)-based evaluation model enhanced with skill optimization and multi-dimensional supervised fine-tuning, to provide human-aligned feedback for MIE. Extensive experiments show that MIEScore achieves state-of-the-art performance in aligning with human preferences and generalizes well across other IEQA datasets. Both the dataset and the model are available at https://github.com/IntMeGroup/MIEScore.

cs.CV↗

arXiv2026

Domain Decoupling Attack: Exploiting the Validation Gap Between Protective DNS and Shared Edge Routing

Network attackers often conceal malicious communication within legitimate Internet traffic. Existing CDN-based evasion techniques rely on SNI--Host inconsistency, insufficient domain ownership verification, or provider-specific routing rewrites, which limit their applicability in modern CDN environments. We identify a validation gap in DNS-based authorization, where permission derived from an allowed domain applies to a shared IP and can be reused to reach another tenant in both CDN and non-CDN shared-hosting environments. This paper presents the Domain Decoupling Attack (DDA), which resolves an allowed domain to obtain permission for a shared edge IP and subsequently connects to the same address while presenting the hidden domain consistently in both TLS SNI and HTTP Host. Measurements of 1,069,048 domains across six continents produce 18,025,068 successful probes and identify exposure rates of 95.8% overall, 99.26% for CDN domains, 92.75% for non-CDN domains, and 97.7% for non-CDN cross-tenant IPs, while laboratory experiments reveal a structural limitation of DNS-bound access control on shared addresses. These results clarify the security risks of DNS-derived IP authorization and support the evaluation and improvement of access-control mechanisms in CDN and non-CDN shared-hosting environments.

cs.CR↗

arXiv2026

Do Uncertainty Signals Help? A Systematic Study of Uncertainty-Aware Decoding with Rollback Mechanisms

Prediction uncertainty is a widely adopted metric for quantifying model confidence, with downstream applications spanning model explanation, data selection, and prediction rollback. Despite its demonstrated utility, the potential of uncertainty quantification to enhance code generation in large language models (LLMs) remains largely underexplored, raising a critical question: to what extent can uncertainty serve as an effective signal for improving LLM-based code generation? To answer this question, we study uncertainty-aware rollback decoding, an inference-time strategy that uses uncertainty signals to identify unreliable generation regions and roll back to earlier valid prefixes without retraining the model. We evaluate this framework on seven code LLMs, five code generation benchmarks, and eight token-level uncertainty signals under a unified decoding setup. Our results show that the complete rollback framework improves over equal-budget restart across the evaluated benchmarks and model settings, with gains of up to 0.26 in pass@1 and 0.35 in AvgTestPassRate on functional code generation benchmarks, and an absolute improvement of up to 6.4\% in Patch-Aligned Safe Rate on Dsec-Python. Among the evaluated signals, information-theoretic measures such as token entropy and negative log-likelihood show the most favorable overall trend, frequently achieving the best or near-best results on standard benchmarks. A component-controlled ablation further shows that feedback-guided rollback provides the main improvement, while uncertainty localization provides an additional gain when checking, budget, rollback, and branch decay are held fixed.

cs.LG↗

arXiv2026

GenSplatCodec: Feed-Forward Gaussian Splatting Compression via One-Step Diffusion

Feed-forward 3D Gaussian Splatting (3DGS) enables scalable scene reconstruction without per-scene optimization, yet produces dense Gaussians that are costly to store and transmit. Existing feed-forward Gaussian compression methods formulate decoding as deterministic representation recovery, which becomes inadequate at low bitrates when high-frequency textures and view-dependent appearance are discarded. Although generative models offer a promising alternative, using them as standalone post-processing decouples generation from the transmitted scene structure, thereby compromising cross-view consistency. To address these limitations, we propose GenSplatCodec, a unified feed-forward Gaussian codec that reformulates low-bitrate Gaussian compression as geometry-guided generative decoding. We present a detail-aware feed-forward Gaussian coding scheme within a dual-stream formulation, where the resulting compact Gaussian structural stream is complemented by a lightweight reference appearance stream. We further introduce a geometry-guided one-step generative decoding approach that jointly exploits decoded structural and appearance cues through hierarchical geometry control to reconstruct high-fidelity and view-consistent novel views. Finally, we develop a three-stage optimization strategy that stabilizes the learning of the unified codec and adapts the generative decoder to codec-derived structural and appearance cues. Extensive experiments across multiple datasets demonstrate that GenSplatCodec consistently achieves superior rate-distortion (RD) performance over existing methods.

cs.CV↗

arXiv2026

Graph-Aware Fuzzing for Graph Database Management Systems

Graph Database Management Systems (GDBMSs) are essential infrastructure for managing interconnected data. Existing GDBMS testing methods primarily rely on differential and metamorphic testing. The result consistency oracles of these methods constrain inputs to queries that are comparable across engines or transformations, leaving single engine runtime failures, such as crashes and memory errors, insufficiently explored. Developing dedicated fuzzers for GDBMSs faces two key challenges: (1) generating valid and structurally diverse queries under complex graph constraints, and (2) guiding exploration to capture topology dependent execution behavior. To address these challenges, we propose GRAF, a black box fuzzing framework for GDBMS query engines. First, GRAF introduces graph context aware query generation based on cascading dependency resolution. It instantiates parameterized Cypher skeletons generated by a Large Language Model (LLM) by jointly resolving labels, relationship types, properties, values, and variable scopes against the active graph state. This process produces structurally diverse queries while eliminating syntactic and semantic violations. Second, GRAF applies five graph specific mutation operators guided by execution state feedback, including execution time, result size, and system status. This feedback steers exploration away from unproductive queries and expensive traversals, while prioritizing local mutations around abnormal executions. We evaluated GRAF against three existing approaches on six widely used GDBMSs. GRAF consistently improves line coverage by 31.6% to 41.1% over the strongest baseline on each target. In 12 hour fuzzing, it triggered 25 unique bugs, compared to six from all baselines combined. Overall, GRAF discovered 34 previously unknown bugs, with 32 confirmed by developers and 23 assigned CVEs.

cs.SE↗

arXiv2026

Robustness and Trade-offs for Code LLMs on Protected Code

Code large language models (LLMs) are increasingly used on software artifacts that may be intentionally obfuscated for intellectual-property protection, reverse-engineering resistance, or controlled access. In reverse engineering and security analysis, deobfuscation is commonly treated as the preprocessing step before downstream analysis or inference, yet its utility for code LLM pipelines has not been systematically validated across models and protection methods. We present an execution-based study of seven code LLMs on protected-code translation and completion, spanning source programs from C++, Go, Java, and JavaScript, five obfuscation methods, and three inference protocols: plain, obfuscated, and deobfuscated. Our results show that direct inference on obfuscated code often matches or exceeds inference on restored code. In this controlled benchmark, higher-capability models such as GPT-4.1 and Qwen3-Coder-30B retain about 90% Pass@1 on obfuscated translation inputs, indicating that explicit restoration is often unnecessary. Same-model restoration does recover some obfuscation-induced failures, but it also degrades many cases that already succeed, with lower-capability models showing the largest net losses. Across settings, model capability is the primary factor, while source language and obfuscation method have secondary but consistent effects. Overall, our findings support model-aware pipeline design and indicate that protected-code workflows should be evaluated primarily with execution-based metrics rather than static similarity alone.

cs.SE↗

arXiv2026

World Narrative Model for Highly Controllable Video Generation: A Paradigm Shift from Pixel Sampling to Physical World Orchestration

The fundamental obstacle to industrial grade video generation is the lack of controllability: existing models treat video as a pixel distribution sampling problem, bypassing the explicit, instance level $4D$ $(3D + T)$ physical world. Consequently, content creators cannot specify geometry, motion, camera parameters, or lighting in a deterministic, quantitative way, leading to the infamous ''gacha'' loop that makes professional content creation prohibitively inefficient and expensive. To address this, we introduce the World Narrative Model (WNM), a paradigm that decouples what to render -- the structured physical narrative -- from how to render -- the pixel generation process. WNM replaces end-to-end black-box sampling with orchestrated $4D$ pre-visualization for media generation. Collaborative agents translate sparse multimodal inputs, including text, reference videos, and sketches, into a fully editable world representation with scene geometry, object layouts, character/animal skeleton motion, trajectories, camera motion, and lighting at quantitative, physically meaningful granularity. This representation acts as a deterministic structural blueprint that drives existing video foundation models, either frozen or lightly adapted, to render final footage, turning the base model into a faithful neural shader. Built on this engine, our human-AI platform supports automatic world generation and pre-visualization aligned with professional filmmaking pipelines, while director consoles enable seamless human refinement. Experiments show that WNM greatly reduces probabilistic ``gacha'' calls and produces videos whose layout, motion, and cinematography closely follow creator intent. The framework is open and modular, allowing each component, such as world representation, control agents, and adapters, to be independently improved. Project website: https://glassroom.sjtu.edu.cn/WNM/.

cs.CV↗

arXiv2026

DTI: Dynamic Trajectory Initialization for Generative Face Video Super-Resolution

As the most perceptually powerful Face Video Super-Resolution (FVSR) method, existing works in Generative FVSR (GFVSR) mainly exploit the generative prior of pretrained diffusion models. However, viewed as full generation, they suffer from fixed sampling and expensive inference costs if without large-scale auxiliary training. Furthermore, an excessive pursuit of generic perceptual metrics often results in low fidelity. To address these issues, we present Dynamic Trajectory Initialization (DTI) paradigm for GFVSR, which reformulates GFVSR as an input-driven directional restoration. With a novel enhancement-and-injection conditioning mechanism for pretrained DiT backbone, fidelity of our model has been significantly improved without compromising perceptual quality. To dynamically set the starting sampling point, we propose a Discriminative Guide (DG) trained via objective Signal-to-Noise Ratio (SNR) alignment. With only minor model adaptation and fine-tuning, our method achieves a SOTA overall performance across diverse metrics and benchmarks. An analysis of relationship between actual comprehensive quality and common metrics is also conducted, which demonstrates the perception-distortion trade-off and that the LPIPS is the most convincing metric in our case.

cs.CV↗

arXiv2026

Mask to Concept: Auto-Promptable SAM3 via Efficient Test-Time Concept Embedding Search for Few-Shot Annotation

Transforming foundation segmentation models from human-prompted tools into auto-promptable annotators is critical for scalable medical data annotation. Current methods commonly depend on external feature matchers or auxiliary networks to automate geometric prompting, but introducing architectural overhead and limiting performance scalability. Although SAM3 natively supports concept segmentation via reusable text prompts, its direct use in medical imaging is hindered by a lack of fine-grained clinical knowledge and the ambiguity of human-written descriptions. In this work, we propose Mask to Concept (M2C), an efficient framework that adapts SAM3 for medical few-shot annotation without external modules, parameter retraining, or manual text engineering. Using only a few labeled images, M2C enables SAM3 to automatically search for transferable visual concepts entirely within its frozen architecture: it initializes a learnable concept embedding, uses it to prompt segmentation, and updates the embedding by gradients of minimizing the concept segmentation error. We further introduce a Hybrid Uncertainty Estimation (HUE) module that calculates the prediction entropy and maps concept predictions back to the box prompts, measuring concept-geometry prompting inconsistency. Highly uncertain samples are flagged actively for human correction, and the corrected masks are then fed back to M2C to continuously search for more precise concept embeddings, forming a self-enhancing annotation loop with minimal expert effort. Experiments on medical segmentation benchmarks show that our method achieves SOTA few-shot segmentation performance and outstanding annotation efficiency, offering a practical and efficient pathway toward scalable medical image labeling. Codes are at https://github.com/Huster-Hq/M2C.

cs.CV↗

arXiv2026

ASAP: A Disaggregated and Asynchronous Inference System for MoE Prefill

Mixture-of-Experts (MoE) models have become the de facto standard for scaling large language models. To maintain computational efficiency, modern MoE serving systems typically employ a hybrid parallelism strategy, combining Data Parallelism (DP) for attention stages with Expert Parallelism (EP) for MoE stages. However, this design necessitates frequent global synchronization barriers between attention DP groups and experts. In online serving, significant variance in request arrival rates and sequence lengths inherently leads to DP imbalance, causing severe synchronization stalls that degrade Time-to-First-Token (TTFT) and system throughput. We present ASAP, an asynchronous inference system specifically designed to accelerate the prefill phase of MoE models. ASAP disaggregates the attention and MoE stages and implements a fully asynchronous execution pipeline. This is achieved through a suite of specialized asynchronous communication primitives and four coordinated optimizations across request scheduling and model execution, which collectively dismantle global synchronization barriers. We implement and evaluate ASAP on CloudMatrix384 super-nodes, demonstrating that it improves SLO-compliant prefill throughput by 90% compared to state-of-the-art synchronous serving solutions.

cs.DC↗

arXiv2026

TEASR: Training-Efficient Any-Step Diffusion Transformer for Real-World Image Super-Resolution

Diffusion models excel in Real-World Image Super-Resolution (Real-ISR) due to their powerful generative priors but suffer from slow iterative sampling. Although existing one-step distillation methods accelerate inference, they typically require auxiliary teacher models that inflate training memory and restrict scalability to large-scale architectures. Furthermore, these fixed-step models lack the flexibility to trade off speed for quality. In this paper, we propose TEASR, a training-efficient any-step diffusion framework for Real-ISR that enables both one-step and multi-step restoration within a unified model. Our key idea is to perform self-adversarial distillation within a single diffusion model, eliminating the need for auxiliary teachers or discriminators. Specifically, we propose a timestep-aware rectification strategy that stabilizes one-step generation across noise levels. These two designs further enables the distillation of 20B-parameter diffusion models on a single GPU, significantly improving training efficiency. Moreover, we introduce a dual-branch diffusion transformer with decoupled timestep condition to separate the current noise state and the denoising target to enhance sampling quality. Extensive experiments demonstrate that TEASR supports seamless any-step sampling and consistently outperforms state-of-the-art methods across multiple datasets.

cs.CV↗

arXiv2026

LL-Bench: Rethinking Low-Level Vision Evaluation in the Era of Large-Scale Generative Models

Large-scale generative models have demonstrated remarkable capabilities across image generation and editing tasks. However, their performance in low-level vision tasks, which require pixel-wise control, remains insufficiently studied. To address this gap, we introduce \textbf{LL-Bench}, a comprehensive \textbf{Benchmark} for evaluating the capabilities of large-scale generative models on \textbf{L}ow-\textbf{L}evel vision tasks. The benchmark comprises 2,469 real-world degraded images covering 16 low-level degradation tasks, and 28,919 restored images produced by 10 state-of-the-art large-scale generative models and 21 conventional restoration models, which are annotated with 152,020 expert-level pairwise human preferences and 28,334 quality scores. Built upon LL-Bench, we present a systematic diagnosis that reveals the performance boundaries and unique failure modes of large-scale generative models across diverse low-level vision tasks, compared with conventional representative restoration approaches. Moreover, we investigate the effectiveness of current quality evaluation metrics on LL-Bench, which exhibit significant discrepancy with human ratings. To better align restored-image quality assessment with human preferences, we further propose \textbf{LL-Score}, an MLLM-based evaluator that captures both restoration quality and hallucination existence. Extensive experiments demonstrate that LL-score not only outperforms existing image quality assessment metrics, but also serves as a promising reward model for training generative models on low-level vision tasks.

cs.CV↗

arXiv2026

Towards Demystifying and Repairing LLM-in-the-Loop Vulnerabilities

Large Language Models (LLMs) have been actively integrated into modern software systems as critical components, introducing a new type of software vulnerability, LLM-in-the-Loop (LiL) vulnerability, in which threats are caused by LLMs. Although some studies have attempted to investigate the impact of LiL vulnerabilities, they have unfortunately failed to clearly distinguish LiL vulnerabilities from conventional ones, leaving the understanding of real-world LiL vulnerabilities an open problem. To address this gap, we first clearly define the scope of LiL vulnerability, and discuss the differences between LiL vulnerabilities and vulnerabilities that exist in LLM systems but are not really caused by LLMs (i.e., LLM-ecosystem vulnerabilities). Then, we construct the first LiL vulnerability dataset, LiLCVE, covering 41 LiL vulnerabilities and 75 LLM-ecosystem vulnerabilities, to facilitate the risk analysis of LLM-integrated software. The analysis of LiLCVE reveals that LiL vulnerabilities have higher severity than LLM-ecosystem vulnerabilities and conventional software vulnerabilities, with 15.5% and 30.3% more critical vulnerabilities, respectively. Furthermore, given the high severity of LiL vulnerabilities and the potential of LLM-based vulnerability repair methods in patching conventional software vulnerabilities. We explore the capabilities of existing widely-used LLM-based methods in repairing vulnerabilities in LiLCVE. Experimental results on 20 agent-model configuration demonstrate that LiL vulnerabilities are far more challenging to fix, with an average decrease of 10.8% Pass@1 rate compared to other types of vulnerabilities. More critically, three categories, Generated Query Execution, Agent Action, and Model Output Rendering, frequently receive 0% repair success rates.

cs.SE↗

arXiv2026

Focal Reward: Balanced Reinforcement Learning under Rubric-Based Rewards

The open-ended generation in LLMs usually requires multi-dimensional rubrics to adequately assess quality and guide the improvement of reinforcement learning. However, a critical dilemma inherent in this training paradigm is the imbalanced reward polarization along different rubric dimensions. Under this bottleneck, even if LLMs achieve relatively high rewards after training, they may still exhibit severe deficiencies in certain dimensions, leading to a direct deterioration in user experience. To address this problem, we propose Focal Reward, a novel objective to automatically balance the training of reinforcement learning under rubric-based rewards. Specifically, we first leverage an inverse reward projection mechanism to estimate the saturation degree of each criterion in the rubric, which forms the basis to calibrate the reward direction. Then, the final objective is designed with an automatically reweighting coefficient for each criterion to achieve the fine-grained balancing. Extensive experiments across three model scales and six benchmarks demonstrate that our Focal Reward method outperforms the strongest static aggregation baseline in all 18 model-benchmark comparisons. Rollout, mechanism, and ablation analyses further show that these gains arise from online, saturation-aware reallocation toward rubrics that still have room for improvement.

cs.LG↗