arXiv subjects

Xinyi Huang

Publications and source records attributed to Xinyi Huang.

At least 19 recordsLinked to original sources

arXiv2026

J-Miner: Recovering Executable Decision Knowledge from Language-Model Classifiers

Large language models can be fine-tuned into specialized classifiers that perform well across diverse text tasks and make complex judgments, but they typically expose only final labels, leaving the decision knowledge acquired through fine-tuning implicit within the model. We study how to mine this internal decision knowledge from a fine-tuned classifier and encode it in an executable representation that can be inspected, validated, and reused beyond the source classifier. We introduce J-Miner, which mines text-level named concepts by aggregating vocabulary-aligned internal signals across layers and token positions, and uses the classifier's own predictions to learn executable decision rules over them. This process distills local internal readouts into an explicit classifier-level knowledge representation. Across multiple classification tasks, J-Miner rules reproduce up to 98.3\% of source-classifier decisions and achieve 6.0--29.5 percentage points higher behavioral fidelity than equally compact rules learned from input words. Further analysis shows that the named concepts reflect internal semantic evidence associated with task decisions, while the learned rules consolidate these distributed signals into inspectable decision structures. The resulting decision knowledge also transfers to lightweight standalone students: using about 1/24 as many parameters as the source classifiers, they reconstruct and execute the representation from raw text while retaining 99.8\% of the source classifiers' mean task accuracy. These findings show that task-specific decision knowledge can be faithfully represented in an explicit, executable form and reused beyond the classifier in which it was learned.

cs.LG↗

arXiv2026

SPA: A SQL-Plan-Aware Reinforcement Learning Framework for Query Rewriting with LLMs

SQL query rewriting is a well-established technique for improving database performance without schema or index changes, yet finding effective rewrites for modern analytical workloads remains difficult: rule-based methods are limited to predefined transformations, while LLM-based approaches often produce rewrites that are semantically valid but compile to equivalent physical plans or degrade runtime performance. We present SPA, a SQL-Plan-Aware reinforcement learning framework that trains LLMs to rewrite queries using physical execution feedback. SPA formulates rewriting as a policy optimization problem and extends GRPO with rewards spanning semantic equivalence, textual rewrite distance, physical-plan divergence, and runtime speedup. To handle reward sparsity across query difficulty, SPA introduces Probability-Gated Adaptive Reward Shaping, a query-level curriculum that unlocks higher-level rewards only once a rollout group achieves sufficient mastery of lower-level objectives, and further improves sample efficiency through on-policy self-improvement by recycling slowdown rewrites from the current policy as targeted training signals. On both IID and OOD workloads, SPA outperforms rule-based and strong LLM baselines in end-to-end runtime, substantially reduces harmful slowdown rewrites, and yields strong tail-latency gains.

cs.DB↗

arXiv2026

Toward Agile and Cooperative LEO Satellite Beam-Hopping Networks: Paradigms, Challenges, and Opportunities

Low-Earth orbit (LEO) satellite beam-hopping (BH) technology is emerging as a promising approach to meet the ever-increasing global connectivity demands, enabling agile, on-demand coverage. LEO satellite BH can address the spatio-temporal non-uniformity of ground user traffic by dynamically allocating capacity and optimizing network performance. Cooperative multi-satellite BH enables joint transmission and interference avoidance to improve received signal quality. This article provides a comprehensive paradigm of BH, detailing its key dimensions, strategies, and architectures. Through exploration of key challenges, including beam pattern design, on-demand scheduling, and interference management, this paper identifies the potential applications of BH, ranging from adaptive capacity allocation for hotspot areas, low-power Internet-of-Things (IoT), delay-sensitive services, to massive connectivity support. Furthermore, a system-level analysis is presented, including key metrics, models of inter-beam and inter-satellite interference, and cooperative joint transmission, and a case study is provided to demonstrate the performance benefits of BH with cooperative transmission. Several promising future research directions are discussed to guide the future development of LEO satellite BH networks.

eess.SP↗

arXiv2026

Structure-Grounded Knowledge Retrieval via Code Dependencies for Multi-Step Data Reasoning

Selecting the right knowledge is critical when using large language models (LLMs) to solve domain-specific data analysis tasks. However, most retrieval-augmented approaches rely primarily on lexical or embedding similarity, which is often a weak proxy for the task-critical knowledge needed for multi-step reasoning. In many such tasks, the relevant knowledge is not merely textually related to the query, but is instead grounded in executable code and the dependency structure through which computations are carried out. To address this mismatch, we propose SGKR (Structure-Grounded Knowledge Retrieval), a retrieval framework that organizes domain knowledge with a graph induced by function-call dependencies. Given a question, SGKR extracts semantic input and output tags, identifies dependency paths connecting them, and constructs a task-relevant subgraph. The associated knowledge and corresponding function implementations are then assembled as a structured context for LLM-based code generation. Experiments on multi-step data analysis benchmarks show that SGKR consistently improves solution correctness over no-retrieval and similarity-based retrieval baselines for both vanilla LLMs and coding agents.

cs.CL↗

arXiv2026

OpenWorldLib: A Unified Codebase and Definition of Advanced World Models

World models have garnered significant attention as a promising research direction in artificial intelligence, yet a clear and unified definition remains lacking. In this paper, we introduce OpenWorldLib, a comprehensive and standardized inference framework for Advanced World Models. Drawing on the evolution of world models, we propose a clear definition: a world model is a model or framework centered on perception, equipped with interaction and long-term memory capabilities, for understanding and predicting the complex world. We further systematically categorize the essential capabilities of world models. Based on this definition, OpenWorldLib integrates models across different tasks within a unified framework, enabling efficient reuse and collaborative inference. Finally, we present additional reflections and analyses on potential future directions for world model research. Code link: https://github.com/OpenDCAI/OpenWorldLib

cs.CV↗

arXiv2026

Silo-Bench: A Scalable Environment for Evaluating Distributed Coordination in Multi-Agent LLM Systems

Large language models are increasingly deployed in multi-agent systems to overcome context limitations by distributing information across agents. Yet whether agents can reliably compute with distributed information, rather than merely exchange it, remains an open question. We introduce SILO-BENCH, a role-agnostic benchmark of 30 algorithmic tasks across three communication complexity levels, evaluating 54 configurations over 1,620 experiments. Our experiments expose a fundamental Communication-Reasoning Gap: agents spontaneously form task-appropriate coordination topologies and exchange information actively, yet systematically fail to synthesize distributed state into correct answers. The failure is localized to the reasoning-integration stage where agents often acquire sufficient information but cannot integrate it. This coordination overhead compounds with scale, eventually eliminating parallelization gains entirely. These findings demonstrate that naively scaling agent count cannot circumvent context limitations, and SILO-BENCH provides a foundation for tracking progress toward genuinely collaborative multi-agent systems. The code is available at https://github.com/jwyjohn/acl26-silo-bench .

cs.MA↗

arXiv2026

Research on World Models Is Not Merely Injecting World Knowledge into Specific Tasks

World models have emerged as a critical frontier in AI research, aiming to enhance large models by infusing them with physical dynamics and world knowledge. The core objective is to enable agents to understand, predict, and interact with complex environments. However, current research landscape remains fragmented, with approaches predominantly focused on injecting world knowledge into isolated tasks, such as visual prediction, 3D estimation, or symbol grounding, rather than establishing a unified definition or framework. While these task-specific integrations yield performance gains, they often lack the systematic coherence required for holistic world understanding. In this paper, we analyze the limitations of such fragmented approaches and propose a unified design specification for world models. We suggest that a robust world model should not be a loose collection of capabilities but a normative framework that integrally incorporates interaction, perception, symbolic reasoning, and spatial representation. This work aims to provide a structured perspective to guide future research toward more general, robust, and principled models of the world.

cs.CV↗

arXiv2025

Backdoor Attacks on Prompt-Driven Video Segmentation Foundation Models

Prompt-driven Video Segmentation Foundation Models (VSFMs), such as SAM2, are increasingly used in applications including autonomous driving and digital pathology, yet their security risks remain underexplored. We study backdoor attacks against VSFMs and show that directly applying classic attacks such as BadNet is largely ineffective, yielding attack success rates (ASR) below 5%. Through gradient-similarity and attention-map analyses, we find that traditional backdoor training fails because clean and triggered samples induce aligned image-encoder gradients, while model attention remains focused on the prompt-specified object rather than the trigger. To address this limitation, we propose BadVSFM, the first backdoor attack framework tailored to prompt-driven VSFMs. BadVSFM uses a two-stage strategy that first learns trigger-specific encoder features and then trains the decoder to map triggered frame prompt representations to an attacker-specified target mask while preserving clean segmentation behavior. Experiments on five VSFMs and two datasets show that BadVSFM achieves strong, controllable backdoor effects across triggers and prompt types with limited clean-performance degradation. Ablations and interpretability analyses validate the necessity of the two-stage design, and five representative defenses remain largely ineffective. Our results reveal a practical and underexplored vulnerability of current VSFMs to backdoor threats.

cs.CV↗

arXiv2025

VABench: A Comprehensive Benchmark for Audio-Video Generation

Recent advances in video generation have been remarkable, enabling models to produce visually compelling videos with synchronized audio. While existing video generation benchmarks provide comprehensive metrics for visual quality, they lack convincing evaluations for audio-video generation, especially for models aiming to generate synchronized audio-video outputs. To address this gap, we introduce VABench, a comprehensive and multi-dimensional benchmark framework designed to systematically evaluate the capabilities of synchronous audio-video generation. VABench encompasses three primary task types: text-to-audio-video (T2AV), image-to-audio-video (I2AV), and stereo audio-video generation. It further establishes two major evaluation modules covering 15 dimensions. These dimensions specifically assess pairwise similarities (text-video, text-audio, video-audio), audio-video synchronization, lip-speech consistency, and carefully curated audio and video question-answering (QA) pairs, among others. Furthermore, VABench covers seven major content categories: animals, human sounds, music, environmental sounds, synchronous physical sounds, complex scenes, and virtual worlds. We provide a systematic analysis and visualization of the evaluation results, aiming to establish a new standard for assessing video generation models with synchronous audio capabilities and to promote the comprehensive advancement of the field.

cs.CV↗

arXiv2025

Consistency requirement of data-driven subgrid-scale modeling in large-eddy simulation

Data-driven subgrid-scale (SGS) modeling in the large-eddy simulations (LES) suffers from the inconsistency between the \textit{a priori} tests and the a posteriori tests, which make training accurate SGS models a difficult task. We study the difference in filtered high-fidelity data and LES to identify the numerical deviation between the two cases, which is a combined impact of commutation error, numerical errors, and error coupling. The impact of the numerical deviation is examined through two SGS model formulations: the eddy-viscosity and the complex nonlinear models. By incorporating numerical deviations into model training, we enhance consistency, stabilize simulations, and improve predictions of energy spectra in a posteriori tests. Our findings highlight that data-driven methods introduce significant nonlinearity and equation coupling, exacerbating inconsistencies compared to non-data-driven approaches. Finally, while the impact of the numerical deviation can be generalized, achieving accurate model predictions necessitates a physically grounded model form and an optimal filter width.

physics.flu-dyn↗

arXiv2025

Work Zones challenge VLM Trajectory Planning: Toward Mitigation and Robust Autonomous Driving

Visual Language Models (VLMs), with powerful multimodal reasoning capabilities, are gradually integrated into autonomous driving by several automobile manufacturers to enhance planning capability in challenging environments. However, the trajectory planning capability of VLMs in work zones, which often include irregular layouts, temporary traffic control, and dynamically changing geometric structures, is still unexplored. To bridge this gap, we conduct the \textit{first} systematic study of VLMs for work zone trajectory planning, revealing that mainstream VLMs fail to generate correct trajectories in $68.0%$ of cases. To better understand these failures, we first identify candidate patterns via subgraph mining and clustering analysis, and then confirm the validity of $8$ common failure patterns through human verification. Building on these findings, we propose REACT-Drive, a trajectory planning framework that integrates VLMs with Retrieval-Augmented Generation (RAG). Specifically, REACT-Drive leverages VLMs to convert prior failure cases into constraint rules and executable trajectory planning code, while RAG retrieves similar patterns in new scenarios to guide trajectory generation. Experimental results on the ROADWork dataset show that REACT-Drive yields a reduction of around $3\times$ in average displacement error relative to VLM baselines under evaluation with Qwen2.5-VL. In addition, REACT-Drive yields the lowest inference time ($0.58$s) compared with other methods such as fine-tuning ($17.90$s). We further conduct experiments using a real vehicle in 15 work zone scenarios in the physical world, demonstrating the strong practicality of REACT-Drive.

cs.RO↗

arXiv2025

Rethinking and Benchmarking Large Language Models for Graph Reasoning

Large Language Models (LLMs) for Graph Reasoning have been extensively studied over the past two years, involving enabling LLMs to understand graph structures and reason on graphs to solve various graph problems, with graph algorithm problems being the most prevalent. Recent studies underscore the potential of LLMs in handling graph reasoning tasks, but their performance is underwhelming. In this work, we point out issues with existing methods and benchmarks, and rethink the direction that LLMs for graph reasoning should strive toward. We find that base models, e.g., GPT-4o-mini, are largely underestimated due to improper reasoning focus. Base models with reasoning focus redirected from replicating graph algorithms to designing them can easily solve most graph reasoning tasks in existing benchmarks. To truly evaluate the graph reasoning capabilities of LLMs, we construct a more challenging GraphAlgorithm benchmark, comprising 239 different graph problems and 3,041 test instances collected from 4 competition platforms. Finally, we introduce a simple and strong baseline Simple-Reasoning-Then-Coding (Simple-RTC)-which guides LLMs to design graph algorithms first and then code to address graph reasoning tasks. Simple-RTC achieves near-perfect accuracy on existing benchmarks and significantly outperforms GPT-4o-mini and all prior methods on the GraphAlgorithm benchmark. This strong baseline encourages further advancements in LLMs for Graph Reasoning in the future.

cs.AI↗

arXiv2025

Multimodal Health Risk Prediction System for Chronic Diseases via Vision-Language Fusion and Large Language Models

With the rising global burden of chronic diseases and the multimodal and heterogeneous clinical data (medical imaging, free-text recordings, wearable sensor streams, etc.), there is an urgent need for a unified multimodal AI framework that can proactively predict individual health risks. We propose VL-RiskFormer, a hierarchical stacked visual-language multimodal Transformer with a large language model (LLM) inference head embedded in its top layer. The system builds on the dual-stream architecture of existing visual-linguistic models (e.g., PaLM-E, LLaVA) with four key innovations: (i) pre-training with cross-modal comparison and fine-grained alignment of radiological images, fundus maps, and wearable device photos with corresponding clinical narratives using momentum update encoders and debiased InfoNCE losses; (ii) a time fusion block that integrates irregular visit sequences into the causal Transformer decoder through adaptive time interval position coding; (iii) a disease ontology map adapter that injects ICD-10 codes into visual and textual channels in layers and infers comorbid patterns with the help of a graph attention mechanism. On the MIMIC-IV longitudinal cohort, VL-RiskFormer achieved an average AUROC of 0.90 with an expected calibration error of 2.7 percent.

cs.AI↗

arXiv2025

SilentLedger: Privacy-Preserving Auditing for Blockchains with Complete Non-Interactivity

Privacy-preserving blockchain systems are essential for protecting transaction data, yet they must also provide auditability that enables auditors to recover participant identities and transaction amounts when warranted. Existing designs often compromise the independence of auditing and transactions, introducing extra interactions that undermine usability and scalability. Moreover, many auditable solutions depend on auditors serving as validators or recording nodes, which introduces risks to both data security and system reliability. To overcome these challenges, we propose SilentLedger, a privacy-preserving transaction system with auditing and complete non-interactivity. To support public verification of authorization, we introduce a renewable anonymous certificate scheme with formal semantics and a rigorous security model. SilentLedger further employs traceable transaction mechanisms constructed from established cryptographic primitives, enabling users to transact without interaction while allowing auditors to audit solely from on-chain data. We formally prove security properties including authenticity, anonymity, confidentiality, and soundness, provide a concrete instantiation, and evaluate performance under a standard 2-2 transaction model. Our implementation and benchmarks demonstrate that SilentLedger achieves superior performance compared with state-of-the-art solutions.

cs.CR↗

arXiv2025

ZPD-SCA: Unveiling the Blind Spots of LLMs in Assessing Students' Cognitive Abilities

Large language models (LLMs) have demonstrated potential in educational applications, yet their capacity to accurately assess the cognitive alignment of reading materials with students' developmental stages remains insufficiently explored. This gap is particularly critical given the foundational educational principle of the Zone of Proximal Development (ZPD), which emphasizes the need to match learning resources with Students' Cognitive Abilities (SCA). Despite the importance of this alignment, there is a notable absence of comprehensive studies investigating LLMs' ability to evaluate reading comprehension difficulty across different student age groups, especially in the context of Chinese language education. To fill this gap, we introduce ZPD-SCA, a novel benchmark specifically designed to assess stage-level Chinese reading comprehension difficulty. The benchmark is annotated by 60 Special Grade teachers, a group that represents the top 0.15% of all in-service teachers nationwide. Experimental results reveal that LLMs perform poorly in zero-shot learning scenarios, with Qwen-max and GLM even falling below the probability of random guessing. When provided with in-context examples, LLMs performance improves substantially, with some models achieving nearly double the accuracy of their zero-shot baselines. These results reveal that LLMs possess emerging abilities to assess reading difficulty, while also exposing limitations in their current training for educationally aligned judgment. Notably, even the best-performing models display systematic directional biases, suggesting difficulties in accurately aligning material difficulty with SCA. Furthermore, significant variations in model performance across different genres underscore the complexity of task. We envision that ZPD-SCA can provide a foundation for evaluating and improving LLMs in cognitively aligned educational applications.

cs.CL↗

arXiv2025

SwitchPatch: Physical Adversarial Attack Strategy with Switchable Adversarial Objectives

Physical adversarial patch (PAP) attacks attach carefully crafted patches to physical objects to manipulate a deployed model. However, existing PAP attacks suffer from several limitations. First, existing patches remain continuously active, which prevents selective targeting of specific attack objectives and compromises stealth. Second, these approaches require target device access or hardware configuration knowledge, and often rely on costly external equipment. To address these limitations, this paper introduces SwitchPatch, a novel physical adversarial attack strategy that employs a physically static adversarial patch yet can be triggered to produce dynamic and controllable attack effects. Unlike existing approaches, SwitchPatch can transition between states through predefined triggers, enabling adaptation to dynamic environments. Moreover, to improve stealth, we design two trigger patterns: one overlapping with the patch and another spatially separated from it. These triggers can be implemented at low cost without target device access or hardware configuration knowledge. We make three contributions. First, we provide theoretical and empirical analysis to establish the feasibility of SwitchPatch and characterize the number of attack objectives it can support. Second, we develop a gradient-based framework for static yet switchable attacks through diverse trigger patterns. Third, we conduct extensive Unmanned Ground Vehicle (UGV) experiments to validate the effectiveness, transferability, and robustness of SwitchPatch.

cs.CR↗

arXiv2025

Evaluation Hallucination in Multi-Round Incomplete Information Lateral-Driven Reasoning Tasks

Multi-round incomplete information tasks are crucial for evaluating the lateral thinking capabilities of large language models (LLMs). Currently, research primarily relies on multiple benchmarks and automated evaluation metrics to assess these abilities. However, our study reveals novel insights into the limitations of existing methods, as they often yield misleading results that fail to uncover key issues, such as shortcut-taking behaviors, rigid patterns, and premature task termination. These issues obscure the true reasoning capabilities of LLMs and undermine the reliability of evaluations. To address these limitations, we propose a refined set of evaluation standards, including inspection of reasoning paths, diversified assessment metrics, and comparative analyses with human performance.

cs.CL↗

arXiv2025

The Eye of Sherlock Holmes: Uncovering User Private Attribute Profiling via Vision-Language Model Agentic Framework

Our research reveals a new privacy risk associated with the vision-language model (VLM) agentic framework: the ability to infer sensitive attributes (e.g., age and health information) and even abstract ones (e.g., personality and social traits) from a set of personal images, which we term "image private attribute profiling." This threat is particularly severe given that modern apps can easily access users' photo albums, and inference from image sets enables models to exploit inter-image relations for more sophisticated profiling. However, two main challenges hinder our understanding of how well VLMs can profile an individual from a few personal photos: (1) the lack of benchmark datasets with multi-image annotations for private attributes, and (2) the limited ability of current multimodal large language models (MLLMs) to infer abstract attributes from large image collections. In this work, we construct PAPI, the largest dataset for studying private attribute profiling in personal images, comprising 2,510 images from 251 individuals with 3,012 annotated privacy attributes. We also propose HolmesEye, a hybrid agentic framework that combines VLMs and LLMs to enhance privacy inference. HolmesEye uses VLMs to extract both intra-image and inter-image information and LLMs to guide the inference process as well as consolidate the results through forensic analysis, overcoming existing limitations in long-context visual reasoning. Experiments reveal that HolmesEye achieves a 10.8% improvement in average accuracy over state-of-the-art baselines and surpasses human-level performance by 15.0% in predicting abstract attributes. This work highlights the urgency of addressing privacy risks in image-based profiling and offers both a new dataset and an advanced framework to guide future research in this area.

cs.CV↗