arXiv subjects

Wenqi Zhang

Publications and source records attributed to Wenqi Zhang.

At least 55 records · Page 3Linked to original sources

arXiv2026

Reinforcement Fine-Tuning for History-Aware Dense Retriever in RAG

Retrieval-augmented generation (RAG) enables large language models (LLMs) to produce evidence-based responses, and its performance hinges on the matching between the retriever and LLMs. Retriever optimization has emerged as an efficient alternative to fine-tuning LLMs. However, existing solutions suffer from objective mismatch between retriever optimization and the goal of RAG pipeline. Reinforcement learning (RL) provides a promising solution to address this limitation, yet applying RL to retriever optimization introduces two fundamental challenges: 1) the deterministic retrieval is incompatible with RL formulations, and 2) state aliasing arises from query-only retrieval in multi-hop reasoning. To address these challenges, we replace deterministic retrieval with stochastic sampling and formulate RAG as a Markov decision process, making retriever optimizable by RL. Further, we incorporate retrieval history into the state at each retrieval step to mitigate state aliasing. Extensive experiments across diverse RAG pipelines, datasets, and retriever scales demonstrate consistent improvements of our approach in RAG performance.

cs.LG↗

arXiv2026

Clustering-Based User Selection in Federated Learning: Metadata Exploitation for 3GPP Networks

Federated learning (FL) enables collaborative model training without sharing raw user data, but conventional simulations often rely on unrealistic data partitioning and current user selection methods ignore data correlation among users. To address these challenges, this paper proposes a metadatadriven FL framework. We first introduce a novel data partition model based on a homogeneous Poisson point process (HPPP), capturing both heterogeneity in data quantity and natural overlap among user datasets. Building on this model, we develop a clustering-based user selection strategy that leverages metadata, such as user location, to reduce data correlation and enhance label diversity across training rounds. Extensive experiments on FMNIST and CIFAR-10 demonstrate that the proposed framework improves model performance, stability, and convergence in non-IID scenarios, while maintaining comparable performance under IID settings. Furthermore, the method shows pronounced advantages when the number of selected users per round is small. These findings highlight the framework's potential for enhancing FL performance in realistic deployments and guiding future standardization.

eess.SP↗

arXiv2026

MirrorGuard: Toward Secure Computer-Use Agents via Simulation-to-Real Reasoning Correction

Large foundation models are integrated into Computer Use Agents (CUAs), enabling autonomous interaction with operating systems through graphical user interfaces (GUIs) to perform complex tasks. This autonomy introduces serious security risks: malicious instructions or visual prompt injections can trigger unsafe reasoning and cause harmful system-level actions. Existing defenses, such as detection-based blocking, prevent damage but often abort tasks prematurely, reducing agent utility. In this paper, we present MirrorGuard, a plug-and-play defense framework that uses simulation-based training to improve CUA security in the real world. To reduce the cost of large-scale training in operating systems, we propose a novel neural-symbolic simulation pipeline, which generates realistic, high-risk GUI interaction trajectories entirely in a text-based simulated environment, which captures unsafe reasoning patterns and potential system hazards without executing real operations. In the simulation environment, MirrorGuard learns to intercept and rectify insecure reasoning chains of CUAs before they produce and execute unsafe actions. In real-world testing, extensive evaluations across diverse benchmarks and CUA architectures show that MirrorGuard significantly mitigates security risks. For instance, on the ByteDance UI-TARS system, it reduces the unsafe rate from 66.5% to 13.0% while maintaining a marginal false refusal rate (FRR). In contrast, the state-of-the-art GuardAgent only achieves a reduction to 53.9% and suffers from a 15.4% higher FRR. Our work proves that simulation-derived defenses can provide robust, real-world protection while maintaining the fundamental utility of the agent. Our code and model are publicly available at https://bmz-q-q.github.io/MirrorGuard/.

cs.AI↗

arXiv2026

Ground What You See: Hallucination-Resistant MLLMs via Caption Feedback, Diversity-Aware Sampling, and Conflict Regularization

While Multimodal Large Language Models (MLLMs) have achieved remarkable success across diverse tasks, their practical deployment is severely hindered by hallucination issues, which become particularly acute during Reinforcement Learning (RL) optimization. This paper systematically analyzes the root causes of hallucinations in MLLMs under RL training, identifying three critical factors: (1) an over-reliance on chained visual reasoning, where inaccurate initial descriptions or redundant information anchor subsequent inferences to incorrect premises; (2) insufficient exploration diversity during policy optimization, leading the model to generate overly confident but erroneous outputs; and (3) destructive conflicts between training samples, where Neural Tangent Kernel (NTK) similarity causes false associations and unstable parameter updates. To address these challenges, we propose a comprehensive framework comprising three core modules. First, we enhance visual localization by introducing dedicated planning and captioning stages before the reasoning phase, employing a quality-based caption reward to ensure accurate initial anchoring. Second, to improve exploration, we categorize samples based on the mean and variance of their reward distributions, prioritizing samples with high variance to focus the model on diverse and informative data. Finally, to mitigate sample interference, we regulate NTK similarity by grouping sample pairs and applying an InfoNCE loss to push overly similar pairs apart and pull dissimilar ones closer, thereby guiding gradient interactions toward a balanced range. Experimental results demonstrate that our proposed method significantly reduces hallucination rates and effectively enhances the inference accuracy of MLLMs.

cs.CV↗

arXiv2026

Fast Collaborative Inference via Distributed Speculative Decoding

Speculative decoding accelerates large language model (LLM) inference by allowing a small draft model to predict multiple future tokens for verification by a larger target model. In AI-native radio access networks (AI-RAN), this enables device-edge collaborative inference but introduces significant uplink overhead, as existing distributed speculative decoding schemes transmit full vocabulary logits at every step. We propose a sparsify-then-sample strategy, Truncated Sparse Logits Transmission (TSLT), which transmits only the logits and indices of a truncated candidate set. We provide theoretical guarantees showing that the acceptance rate is preserved under TSLT. TSLT is further extended to multi-candidate case, where multiple draft candidates per step increase acceptance probability. Experiments show that TSLT significantly reduces uplink communication while maintaining end-to-end inference latency and model quality, demonstrating its effectiveness for scalable, communication-efficient distributed LLM inference in future AI-RAN systems.

eess.SP↗

arXiv2026

Towards a Theoretical Framework for Robust Node Deployment in Cooperative ISAC Networks

This paper investigates node deployment strategies for robust multi-node cooperative localization in integrated sensing and communication (ISAC) networks.We first analyze how steering vector correlation across different positions affects localization performance and introduce a novel distance-weighted correlation metric to characterize this effect. Building upon this insight, we propose a deployment optimization framework that minimizes the maximum weighted steering vector correlation by optimizing simultaneously node positions and array orientations, thereby enhancing worst-case network robustness. Then, a genetic algorithm (GA) is developed to solve this min-max optimization, yielding optimized node positions and array orientations. Extensive simulations using both multiple signal classification (MUSIC) and neural-network (NN)-based localization validate the effectiveness of the proposed methods, demonstrating significant improvements in robust localization performance.

eess.SP↗

arXiv2025

No Vision, No Wearables: 5G-based 2D Human Pose Recognition with Integrated Sensing and Communications

With the increasing maturity of contactless human pose recognition (HPR) technology, indoor interactive applications have raised higher demands for natural, controller-free interaction methods. However, current mainstream HPR solutions relying on vision or radio-frequency (RF) (including WiFi, radar) still face various challenges in practical deployment, such as privacy concerns, susceptibility to occlusion, dedicated equipment and functions, and limited sensing resolution and range. 5G-based integrated sensing and communication (ISAC) technology, by merging communication and sensing functions, offers a new approach to address these challenges in contactless HPR. We propose a practical 5G-based ISAC system capable of inferring 2D HPR from uplink sounding reference signals (SRS). Specifically, rich features are extracted from multiple domains and employ an encoder to achieve unified alignment and representation in a latent space. Subsequently, low-dimensional features are fused to output the human pose state. Experimental results demonstrate that in typical indoor environments, our proposed 5G-based ISAC HPR system significantly outperforms current mainstream baseline solutions in HPR performance, providing a solid technical foundation for universal human-computer interaction.

eess.SP↗

arXiv2025

Instance Communication System for Intelligent Connected Vehicles: Bridging the Gap from Semantic to Instance-Level Transmission

Intelligent Connected Vehicles (ICVs) rely on high-speed data transmission for efficient and safety-critical services. However, the scarcity of wireless resources limits the capabilities of ICVs. Semantic Communication (SemCom) systems can alleviate this issue by extracting and transmitting task-relevant information, termed semantic information, instead of the entire raw data. Despite this, we reveal that residual redundancy persists within SemCom systems, where not all instances under the same semantic category are equally critical for downstream tasks. To tackle this issue, we introduce Instance Communication (InsCom), which elevates communication from the semantic level to the instance level for ICVs. Specifically, InsCom uses a scene graph generation model to identify all image instances and analyze their inter-relationships, thus distinguishing between semantically identical instances. Additionally, it applies user-configurable, task-critical criteria based on subject semantics and relation-object pairs to filter recognized instances. Consequently, by transmitting only task-critical instances, InsCom significantly reduces data redundancy, substantially enhancing transmission efficiency within limited wireless resources. Evaluations across various datasets and wireless channel conditions show that InsCom achieves a data volume reduction of over 7.82 times and a quality improvement ranging from 1.75 to 14.03 dB compared to the state-of-the-art SemCom systems.

eess.SP↗

arXiv2025

Tool-RoCo: An Agent-as-Tool Self-organization Large Language Model Benchmark in Multi-robot Cooperation

This study proposes Tool-RoCo, a novel benchmark for evaluating large language models (LLMs) in long-term multi-agent cooperation based on RoCo, a multi-robot cooperative benchmark. Recent research on LLM-based multi-agent systems has relied on predefined orchestration, while ignoring agent autonomy. Tool-RoCo treats other agents as tools and introduces cooperative tools, leveraging tool usage to evaluate multi-agent cooperation and self-organization. Tool usage means that each agent (LLM) selects a tool from a candidate set based on the current state, receives feedback, and adjusts its selection in subsequent rounds. To evaluate different autonomy levels, we propose four LLM paradigms: (1) centralized cooperation, where a single LLM allocates tools to all agents; (2) centralized self-organization, where a central LLM autonomously activates agents while keeping others inactive; (3) decentralized cooperation, where each agent has its own LLM and calls tools based on local information; and (4) self-organization, where a randomly chosen initial agent can request collaboration, activating additional agents via tool calls. Tool-RoCo includes three multi-robot tasks, SORT, PACK, and CABINET, to measure format and parameter accuracy and agent coordination through tool usage. The results using several LLMs showed that cooperative tools accounted for only 7.09% of all tools, indicating that LLM-based agents rarely invoked others as assistants. Moreover, activation tools accounted for 96.42%, suggesting that current LLMs tend to maintain active agents while seldom deactivating them for adaptive coordination. Tool-RoCo provides a systematic benchmark to evaluate LLM autonomy and cooperation in multi-agent tasks. Code and Demo: https://github.com/ColaZhang22/Tool-Roco

cs.MA↗

arXiv2025

Mind the Gap: Bridging Thought Leap for Improved Chain-of-Thought Tuning

Large language models (LLMs) have achieved remarkable progress on mathematical tasks through Chain-of-Thought (CoT) reasoning. However, existing mathematical CoT datasets often suffer from Thought Leaps due to experts omitting intermediate steps, which negatively impacts model learning and generalization. We propose the CoT Thought Leap Bridge Task, which aims to automatically detect leaps and generate missing intermediate reasoning steps to restore the completeness and coherence of CoT. To facilitate this, we constructed a specialized training dataset called ScaleQM+, based on the structured ScaleQuestMath dataset, and trained CoT-Bridge to bridge thought leaps. Through comprehensive experiments on mathematical reasoning benchmarks, we demonstrate that models fine-tuned on bridged datasets consistently outperform those trained on original datasets, with improvements of up to +5.87% on NuminaMath. Our approach effectively enhances distilled data (+3.02%) and provides better starting points for reinforcement learning (+3.1%), functioning as a plug-and-play module compatible with existing optimization techniques. Furthermore, CoT-Bridge demonstrate improved generalization to out-of-domain logical reasoning tasks, confirming that enhancing reasoning completeness yields broadly applicable benefits.

cs.CL↗

arXiv2025

Let LRMs Break Free from Overthinking via Self-Braking Tuning

Large reasoning models (LRMs), such as OpenAI o1 and DeepSeek-R1, have significantly enhanced their reasoning capabilities by generating longer chains of thought, demonstrating outstanding performance across a variety of tasks. However, this performance gain comes at the cost of a substantial increase in redundant reasoning during the generation process, leading to high computational overhead and exacerbating the issue of overthinking. Although numerous existing approaches aim to address the problem of overthinking, they often rely on external interventions. In this paper, we propose a novel framework, Self-Braking Tuning (SBT), which tackles overthinking from the perspective of allowing the model to regulate its own reasoning process, thus eliminating the reliance on external control mechanisms. We construct a set of overthinking identification metrics based on standard answers and design a systematic method to detect redundant reasoning. This method accurately identifies unnecessary steps within the reasoning trajectory and generates training signals for learning self-regulation behaviors. Building on this foundation, we develop a complete strategy for constructing data with adaptive reasoning lengths and introduce an innovative braking prompt mechanism that enables the model to naturally learn when to terminate reasoning at an appropriate point. Experiments across mathematical benchmarks (AIME, AMC, MATH500, GSM8K) demonstrate that our method reduces token consumption by up to 60% while maintaining comparable accuracy to unconstrained models.

cs.CL↗

arXiv2025

OmniEduBench: A Comprehensive Chinese Benchmark for Evaluating Large Language Models in Education

With the rapid development of large language models (LLMs), various LLM-based works have been widely applied in educational fields. However, most existing LLMs and their benchmarks focus primarily on the knowledge dimension, largely neglecting the evaluation of cultivation capabilities that are essential for real-world educational scenarios. Additionally, current benchmarks are often limited to a single subject or question type, lacking sufficient diversity. This issue is particularly prominent within the Chinese context. To address this gap, we introduce OmniEduBench, a comprehensive Chinese educational benchmark. OmniEduBench consists of 24.602K high-quality question-answer pairs. The data is meticulously divided into two core dimensions: the knowledge dimension and the cultivation dimension, which contain 18.121K and 6.481K entries, respectively. Each dimension is further subdivided into 6 fine-grained categories, covering a total of 61 different subjects (41 in the knowledge and 20 in the cultivation). Furthermore, the dataset features a rich variety of question formats, including 11 common exam question types, providing a solid foundation for comprehensively evaluating LLMs' capabilities in education. Extensive experiments on 11 mainstream open-source and closed-source LLMs reveal a clear performance gap. In the knowledge dimension, only Gemini-2.5 Pro surpassed 60\% accuracy, while in the cultivation dimension, the best-performing model, QWQ, still trailed human intelligence by nearly 30\%. These results highlight the substantial room for improvement and underscore the challenges of applying LLMs in education.

cs.CL↗

arXiv2025

Active Confusion Expression in Large Language Models: Leveraging World Models toward Better Social Reasoning

While large language models (LLMs) excel in mathematical and code reasoning, we observe they struggle with social reasoning tasks, exhibiting cognitive confusion, logical inconsistencies, and conflation between objective world states and subjective belief states. Through deteiled analysis of DeepSeek-R1's reasoning trajectories, we find that LLMs frequently encounter reasoning impasses and tend to output contradictory terms like "tricky" and "confused" when processing scenarios with multiple participants and timelines, leading to erroneous reasoning or infinite loops. The core issue is their inability to disentangle objective reality from agents' subjective beliefs. To address this, we propose an adaptive world model-enhanced reasoning mechanism that constructs a dynamic textual world model to track entity states and temporal sequences. It dynamically monitors reasoning trajectories for confusion indicators and promptly intervenes by providing clear world state descriptions, helping models navigate through cognitive dilemmas. The mechanism mimics how humans use implicit world models to distinguish between external events and internal beliefs. Evaluations on three social benchmarks demonstrate significant improvements in accuracy (e.g., +10% in Hi-ToM) while reducing computational costs (up to 33.8% token reduction), offering a simple yet effective solution for deploying LLMs in social contexts.

cs.CL↗

arXiv2025

SpatialLadder: Progressive Training for Spatial Reasoning in Vision-Language Models

Spatial reasoning remains a fundamental challenge for Vision-Language Models (VLMs), with current approaches struggling to achieve robust performance despite recent advances. We identify that this limitation stems from a critical gap: existing methods attempt to learn spatial reasoning directly without establishing the hierarchical foundations of perception and understanding. To address this challenge, we present a comprehensive methodology for building spatial intelligence progressively. We introduce SpatialLadder-26k, a multimodal dataset containing 26,610 samples spanning object localization, single image, multi-view, and video spatial reasoning tasks, constructed through a standardized pipeline that ensures systematic coverage across modalities. Building on this dataset, we design a three-stage progressive training framework that (1) establishes spatial perception through object localization, (2) develops spatial understanding through multi-dimensional spatial tasks, and (3) strengthens complex reasoning via reinforcement learning with verifiable rewards. This approach yields SpatialLadder, a 3B-parameter model that achieves state-of-the-art performance on spatial reasoning benchmarks, with 23.4% average improvement over the base model, surpassing GPT-4o by 20.8% and Gemini-2.0-Flash by 10.1%. Notably, SpatialLadder maintains strong generalization with 7.2% improvement on out-of-domain benchmarks, demonstrating that progressive training from perception to reasoning is essential for robust spatial intelligence.

cs.CV↗

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

RG-Attn: Radian Glue Attention for Multi-modality Multi-agent Cooperative Perception

Cooperative perception enhances autonomous driving by leveraging Vehicle-to-Everything (V2X) communication for multi-agent sensor fusion. However, most existing methods rely on single-modal data sharing, limiting fusion performance, particularly in heterogeneous sensor settings involving both LiDAR and cameras across vehicles and roadside units (RSUs). To address this, we propose Radian Glue Attention (RG-Attn), a lightweight and generalizable cross-modal fusion module that unifies intra-agent and inter-agent fusion via transformation-based coordinate alignment and a unified sampling/inversion strategy. RG-Attn efficiently aligns features through a radian-based attention constraint, operating column-wise on geometrically consistent regions to reduce overhead and preserve spatial coherence, thereby enabling accurate and robust fusion. Building upon RG-Attn, we propose three cooperative architectures. The first, Paint-To-Puzzle (PTP), prioritizes communication efficiency but assumes all agents have LiDAR, optionally paired with cameras. The second, Co-Sketching-Co-Coloring (CoS-CoCo), offers maximal flexibility, supporting any sensor setup (e.g., LiDAR-only, camera-only, or both) and enabling strong cross-modal generalization for real-world deployment. The third, Pyramid-RG-Attn Fusion (PRGAF), aims for peak detection accuracy with the highest computational overhead. Extensive evaluations on simulated and real-world datasets show our framework delivers state-of-the-art detection accuracy with high flexibility and efficiency. GitHub Link: https://github.com/LantaoLi/RG-Attn

cs.RO↗

arXiv2025

Anomaly Detection in Industrial Control Systems Based on Cross-Domain Representation Learning

Industrial control systems (ICSs) are widely used in industry, and their security and stability are very important. Once the ICS is attacked, it may cause serious damage. Therefore, it is very important to detect anomalies in ICSs. ICS can monitor and manage physical devices remotely using communication networks. The existing anomaly detection approaches mainly focus on analyzing the security of network traffic or sensor data. However, the behaviors of different domains (e.g., network traffic and sensor physical status) of ICSs are correlated, so it is difficult to comprehensively identify anomalies by analyzing only a single domain. In this paper, an anomaly detection approach based on cross-domain representation learning in ICSs is proposed, which can learn the joint features of multi-domain behaviors and detect anomalies within different domains. After constructing a cross-domain graph that can represent the behaviors of multiple domains in ICSs, our approach can learn the joint features of them by leveraging graph neural networks. Since anomalies behave differently in different domains, we leverage a multi-task learning approach to identify anomalies in different domains separately and perform joint training. The experimental results show that the performance of our approach is better than existing approaches for identifying anomalies in ICSs.

cs.CR↗

arXiv2025

Know What, Know Why: Semantic Hazard Communication for Intelligent V2X Systems

In current vehicle-to-everything (V2X) communication systems, roadside units (RSUs) broadcast brief warning messages that alert nearby vehicles to avoid potential hazards. However, these messages lack contextual information on why a warning is issued, leading to excessive caution or inefficient driving behaviors. To avoid such a situation, we propose a semantic-enhanced and explainable V2X (SEE-V2X) system. In the proposed system, RSUs equipped with smart cameras detect obstructions and transmit context-aware messages to vehicles. By understanding both what the hazard is and why it occurs, drivers can make more intelligent decisions based on their specific driving situation. Furthermore, through a real-field demonstration, we show the new "see-through" feature in the proposed system, which enables drivers to visualize hidden pedestrians behind obstacles. We also perform simulations to compare traditional V2X with SEE-V2X under different traffic conditions. The results show that SEE-V2X significantly improves traffic efficiency and reduces unnecessary deceleration.

eess.SP↗