arXiv subjects

Jun Zhang

Publications and source records attributed to Jun Zhang.

At least 361 records · Page 20Linked to original sources

arXiv2025

VideoCap-R1: Enhancing MLLMs for Video Captioning via Structured Thinking

While recent advances in reinforcement learning have significantly enhanced reasoning capabilities in large language models (LLMs), these techniques remain underexplored in multi-modal LLMs for video captioning. This paper presents the first systematic investigation of GRPO-based RL post-training for video MLLMs, with the goal of enhancing video MLLMs' capability of describing actions in videos. Specifically, we develop the VideoCap-R1, which is prompted to first perform structured thinking that analyzes video subjects with their attributes and actions before generating complete captions, supported by two specialized reward mechanisms: a LLM-free think scorer evaluating the structured thinking quality and a LLM-assisted caption scorer assessing the output quality. The RL training framework effectively establishes the connection between structured reasoning and comprehensive description generation, enabling the model to produce captions with more accurate actions. Our experiments demonstrate that VideoCap-R1 achieves substantial improvements over the Qwen2VL-7B baseline using limited samples (1.5k) across multiple video caption benchmarks (DREAM1K: +4.4 event F1, VDC: +4.2 Acc, CAREBENCH: +3.1 action F1, +6.9 object F1) while consistently outperforming the SFT-trained counterparts, confirming GRPO's superiority in enhancing MLLMs' captioning capabilities.

cs.CV↗

arXiv2025

CityBench: Evaluating the Capabilities of Large Language Models for Urban Tasks

As large language models (LLMs) continue to advance and gain widespread use, establishing systematic and reliable evaluation methodologies for LLMs and vision-language models (VLMs) has become essential to ensure their real-world effectiveness and reliability. There have been some early explorations about the usability of LLMs for limited urban tasks, but a systematic and scalable evaluation benchmark is still lacking. The challenge in constructing a systematic evaluation benchmark for urban research lies in the diversity of urban data, the complexity of application scenarios and the highly dynamic nature of the urban environment. In this paper, we design \textit{CityBench}, an interactive simulator based evaluation platform, as the first systematic benchmark for evaluating the capabilities of LLMs for diverse tasks in urban research. First, we build \textit{CityData} to integrate the diverse urban data and \textit{CitySimu} to simulate fine-grained urban dynamics. Based on \textit{CityData} and \textit{CitySimu}, we design 8 representative urban tasks in 2 categories of perception-understanding and decision-making as the \textit{CityBench}. With extensive results from 30 well-known LLMs and VLMs in 13 cities around the world, we find that advanced LLMs and VLMs can achieve competitive performance in diverse urban tasks requiring commonsense and semantic understanding abilities, e.g., understanding the human dynamics and semantic inference of urban images. Meanwhile, they fail to solve the challenging urban tasks requiring professional knowledge and high-level numerical abilities, e.g., geospatial prediction and traffic control task.

cs.AI↗

arXiv2025

HarmoniCa: Harmonizing Training and Inference for Better Feature Caching in Diffusion Transformer Acceleration

Diffusion Transformers (DiTs) excel in generative tasks but face practical deployment challenges due to high inference costs. Feature caching, which stores and retrieves redundant computations, offers the potential for acceleration. Existing learning-based caching, though adaptive, overlooks the impact of the prior timestep. It also suffers from misaligned objectives--aligned predicted noise vs. high-quality images--between training and inference. These two discrepancies compromise both performance and efficiency. To this end, we harmonize training and inference with a novel learning-based caching framework dubbed HarmoniCa. It first incorporates Step-Wise Denoising Training (SDT) to ensure the continuity of the denoising process, where prior steps can be leveraged. In addition, an Image Error Proxy-Guided Objective (IEPO) is applied to balance image quality against cache utilization through an efficient proxy to approximate the image error. Extensive experiments across $8$ models, $4$ samplers, and resolutions from $256\times256$ to $2K$ demonstrate superior performance and speedup of our framework. For instance, it achieves over $40\%$ latency reduction (i.e., $2.07\times$ theoretical speedup) and improved performance on PixArt-$α$. Remarkably, our image-free approach reduces training time by $25\%$ compared with the previous method. Our code is available at https://github.com/ModelTC/HarmoniCa.

cs.CV↗

arXiv2025

SEAR: A Multimodal Dataset for Analyzing AR-LLM-Driven Social Engineering Behaviors

The SEAR Dataset is a novel multimodal resource designed to study the emerging threat of social engineering (SE) attacks orchestrated through augmented reality (AR) and multimodal large language models (LLMs). This dataset captures 180 annotated conversations across 60 participants in simulated adversarial scenarios, including meetings, classes and networking events. It comprises synchronized AR-captured visual/audio cues (e.g., facial expressions, vocal tones), environmental context, and curated social media profiles, alongside subjective metrics such as trust ratings and susceptibility assessments. Key findings reveal SEAR's alarming efficacy in eliciting compliance (e.g., 93.3% phishing link clicks, 85% call acceptance) and hijacking trust (76.7% post-interaction trust surge). The dataset supports research in detecting AR-driven SE attacks, designing defensive frameworks, and understanding multimodal adversarial manipulation. Rigorous ethical safeguards, including anonymization and IRB compliance, ensure responsible use. The SEAR dataset is available at https://github.com/INSLabCN/SEAR-Dataset.

cs.AI↗

arXiv2025

The experimental determination of exchange mass terms in surface states on both terminations of MnBi4Te7

The intrinsic antiferromagnetic topological insulators in the Mn-Bi-Te family, composed of superlattice-like MnBi2Te4/(Bi2Te3)n (n = 0, 1, 2, 3...) layered structure, present intriguing states of matter such as quantum anomalous Hall effect and the axion insulator. However, the surface state gap, which is the prerequisite for the observation of these states, remains elusive. Here by molecular beam epitaxy, we obtain two types of MnBi4Te7 films with the exclusive Bi2Te3 (BT) or MnBi2Te4 (MBT) terminations. By scanning tunneling spectroscopy, the mass terms in the surface states are identified on both surface terminations. Experimental results reveal the existence of a hybridization gap of approximately 23 meV in surface states on the BT termination. This gap comes from the hybridization between the surface states and the spin-split states in the adjacent MBT layer. On the MBT termination, an exchange mass term of about 30 meV in surface states is identified by taking magnetic-field-dependent Landau level spectra as well as theoretical simulations. In addition, the mass term varies with the field in the film with a heavy BiMn doping level in the Mn layers. These findings demonstrate the existence of mass terms in surface states on both types of terminations in our epitaxial MnBi4Te7 films investigated by local probes.

cond-mat.mtrl-sci↗

arXiv2025

Towards Real-time Structural Dynamics Simulation with Graph-based Digital Twin Modelling

Precise and timely simulation of a structure's dynamic behavior is crucial for evaluating its performance and assessing its health status. Traditional numerical methods are often limited by high computational costs and low efficiency, while deep learning approaches offer a promising alternative. However, these data-driven methods still face challenges, such as limited physical interpretability and difficulty in adapting to diverse structural configurations. To address these issues, this study proposes a graph-based digital twin modelling (GDTM) framework to simulate structural dynamic responses across various spatial topologies. In this framework, the adjacency matrix explicitly represents the spatial relationships between structural vertices, enhancing the model's physical interpretability. The effectiveness of the proposed framework was validated through comprehensive numerical and experimental studies. The results demonstrate that the framework accurately simulated structural dynamics across different topological configurations, with Normalized Mean-Squared Error (NMSE) values consistently below 0.005 in numerical simulations and 0.0015 in experimental validations. Furthermore, the framework achieved over 80-fold improvements in computational efficiency compared to traditional finite element methods (FEM). This research promotes the practical application of graph-based structural dynamics modelling, which has the potential to significantly advance structural performance evaluation and health monitoring.

cs.CE↗

arXiv2025

Enhancing Large-Scale AI Training Efficiency: The C4 Solution for Real-Time Anomaly Detection and Communication Optimization

The emergence of Large Language Models (LLMs) has necessitated the adoption of distributed training techniques, involving the deployment of thousands of GPUs to train a single model. Unfortunately, the efficiency of large-scale distributed training systems is often suboptimal due to the increased likelihood of hardware errors in high-end GPU products and the heightened risk of network traffic collisions. Moreover, any local hardware failure can disrupt training tasks, and the inability to swiftly identify faulty components leads to a significant waste of GPU resources. And, prolonged communication due to traffic collisions can substantially increase GPU waiting times. To address these challenges, we propose a communication-driven solution, namely the C4. The key insights of C4 are twofold. First, the load in distributed training exhibits homogeneous characteristics and is divided into iterations through periodic synchronization, therefore hardware anomalies would incur certain syndrome in collective communication. By leveraging this feature, C4 can rapidly identify the faulty components, swiftly isolate the anomaly, and restart the task, thereby avoiding resource wastage caused by delays in anomaly detection. Second, the predictable communication model of collective communication, involving a limited number of long-lived flows, allows C4 to efficiently execute traffic planning, substantially reducing bandwidth competition among these flows. The C4 has been extensively deployed across real-world production systems in a hyperscale cloud provider, yielding a significant improvement in system efficiency, from 30% to 45%. This enhancement is attributed to a 30% reduction in error-induced overhead and a 15% reduction in communication costs.

cs.DC↗

arXiv2025

Prot2Chat: Protein LLM with Early-Fusion of Text, Sequence and Structure

Motivation: Proteins are of great significance in living organisms. However, understanding their functions encounters numerous challenges, such as insufficient integration of multimodal information, a large number of training parameters, limited flexibility of classification-based methods, and the lack of systematic evaluation metrics for protein Q&A systems. To tackle these issues, we propose the Prot2Chat framework. Results: We modified ProteinMPNN to encode protein sequence and structural information in a unified way. We used a large language model (LLM) to encode questions into vectors and developed a protein-text adapter to compress protein information into virtual tokens based on these vectors, achieving the early fusion of text and protein information. Finally, the same LLM reads the virtual tokens and the questions to generate answers. To optimize training efficiency, we froze the encoder and employed Low-Rank Adaptation (LoRA) techniques for the LLM. Experiments on two datasets show that both automated metrics and expert evaluations demonstrate the superior performance of our model, and zero-shot prediction results highlight its generalization ability. The models and codes are available at https://github.com/ wangzc1233/Prot2Chat. Contact: zqcao@suda.edu.cn or wangzc025@163.com Key words: Protein Q&A, Early-Fusion, LLM

cs.LG↗

arXiv2025

Selective Invocation for Multilingual ASR: A Cost-effective Approach Adapting to Speech Recognition Difficulty

Although multilingual automatic speech recognition (ASR) systems have significantly advanced, enabling a single model to handle multiple languages, inherent linguistic differences and data imbalances challenge SOTA performance across all languages. While language identification (LID) models can route speech to the appropriate ASR model, they incur high costs from invoking SOTA commercial models and suffer from inaccuracies due to misclassification. To overcome these, we propose SIMA, a selective invocation for multilingual ASR that adapts to the difficulty level of the input speech. Built on a spoken large language model (SLLM), SIMA evaluates whether the input is simple enough for direct transcription or requires the invocation of a SOTA ASR model. Our approach reduces word error rates by 18.7% compared to the SLLM and halves invocation costs compared to LID-based methods. Tests on three datasets show that SIMA is a scalable, cost-effective solution for multilingual ASR applications.

cs.SD↗

arXiv2025

Toward Near-Space Communication Network in the 6G and Beyond Era

Near-space communication network (NS-ComNet), as an indispensable component of sixth-generation (6G) and beyond mobile communication systems and the space-air-ground-sea integrated network (SAGSIN), demonstrates unique advantages in wide-area coverage, long-endurance high-altitude operation, and highly flexible deployment. This paper presents a comprehensive review of NS-ComNet for 6G and beyond era. Specifically, by contrasting satellite, low-altitude unmanned-aerial-vehicle (UAV), and terrestrial communications, we first elucidate the background and motivation for integrating NS-ComNet into 6G network architectures. Subsequently, we review the developmental status of near-space platforms, including high-altitude balloons, solar-powered UAVs, and stratospheric airships, and analyze critical challenges faced by NS-ComNet. To address these challenges, the research focuses on key enabling technologies such as topology design, resource and handover management, multi-objective joint optimization, etc., with particular emphasis on artificial intelligence techniques for NS-ComNet. Finally, envisioning future intelligent collaborative networks that integrate NS-ComNet with satellite-UAV-terrestrial systems, we explore promising directions. This paper aims to provide technical insights and research foundations for the systematic construction of NS-ComNet and its deep deployment in the 6G and beyond era.

eess.SP↗

arXiv2025

SALMONN-omni: A Standalone Speech LLM without Codec Injection for Full-duplex Conversation

In order to enable fluid and natural human-machine speech interaction, existing full-duplex conversational systems often adopt modular architectures with auxiliary components such as voice activity detectors, interrupters, conversation state predictors, or multiple LLMs. These systems, however, suffer from error accumulation across modules and struggle with key challenges such as context-dependent barge-in and echo cancellation. Recent approaches, most notably Moshi, simplify the pipeline by injecting audio codecs into the token space of a single LLM. However, such methods still incur significant performance degradation when operating on the speech rather than text modality. In this paper, we introduce SALMONN-omni, the first single, standalone full-duplex speech LLM that operates without audio codecs in its token space. It features a novel dynamic thinking mechanism within the LLM backbone, enabling the model to learn when to transition between speaking and listening states. Experiments on widely used benchmarks for spoken question answering and open-domain dialogue show that SALMONN-omni achieves at least 30\% relative performance improvement over existing open-source full-duplex models and performs highly competitively to half-duplex and turn-based systems, despite using substantially less training data. Moreover, SALMONN-omni demonstrates strong performance in complex conversational scenarios, including turn-taking, backchanneling, echo cancellation and context-dependent barge-in, with further improvements achieved through reinforcement learning. Some demo conversations between user and SALMONN-omni are provided in the following repository https://github.com/bytedance/SALMONN.

cs.CL↗

arXiv2025

Recurrent Jetlets Associated with the Disappearance of a Satellite Spot

Recurrent small-scale eruptions are fascinating phenomena in the solar atmosphere. However, their underlying physical mechanisms remain unclear. On 2021 May 23, five recurrent jetlets (J1-J5) were observed continuously ejecting from a satellite spot located at the north edge of AR 12824. Using high-resolution, multi-wavelength data from NVST, SDO, and IRIS, we investigate the physical characteristics of these jetlets and their relationship with the satellite spot. The widths of these jetlets range from 1300 to 2900 km, their lifetimes range span 3 to 10 minutes, and their projection speeds vary from 152.8 to 406.0 km s$^{-1}$. During the eruptions, the satellite spot moved northwest at a low speed of 376 $\pm$ 12 m s$^{-1}$. Its area gradually decreased due to magnetic cancellation with surrounding positive magnetic field, resulting in an average cancellation rate of 1.3$\times$10$^{18}$ Mx hr$^{-1}$. Dark lanes that separated from the satellite spot and small pores were observed to move toward nearby these features or dark lanes with opposite polarities, eventually disappearing during the magnetic cancellation process. J4 was driven by an eruption of a micro-filament. Spectral observations revealed a redshift on the right side of J4 and a blueshift on the left side of its base, suggesting a counterclockwise rotation. The horizontal magnetic field of the satellite spot consistently exhibited a vortex structure throughout its evolution until it vanished. The nonlinear force-free field extrapolation confirms that the satellite spot serves as one footpoint of a mini-flux rope. These observations reveal that these jetlets might result from three-dimensional null-point magnetic reconnection, initiated by the continuous eruption of a mini-flux-rope or multiple mini-flux-ropes, driven by sustained magnetic cancellation.

astro-ph.SR↗

arXiv2025

FullStack Bench: Evaluating LLMs as Full Stack Coders

As the capabilities of code large language models (LLMs) continue to expand, their applications across diverse code intelligence domains are rapidly increasing. However, most existing datasets only evaluate limited application domains. To address this gap, we have developed a comprehensive code evaluation dataset FullStack Bench focusing on full-stack programming, which encompasses a wide range of application domains (e.g., basic programming, data analysis, software engineering, mathematics, and machine learning). Besides, to assess multilingual programming capabilities, in FullStack Bench, we design real-world instructions and corresponding unit test cases from 16 widely-used programming languages to reflect real-world usage scenarios rather than simple translations. Moreover, we also release an effective code sandbox execution tool (i.e., SandboxFusion) supporting various programming languages and packages to evaluate the performance of our FullStack Bench efficiently. Comprehensive experimental results on our FullStack Bench demonstrate the necessity and effectiveness of our FullStack Bench and SandboxFusion.

cs.AI↗

arXiv2025

Fluid Antenna-Assisted MU-MIMO Systems with Decentralized Baseband Processing

The fluid antenna system (FAS) has emerged as a disruptive technology, offering unprecedented degrees of freedom (DoF) for wireless communication systems. However, optimizing fluid antenna (FA) positions entails significant computational costs, especially when the number of FAs is large. To address this challenge, we introduce a decentralized baseband processing (DBP) architecture to FAS, which partitions the FA array into clusters and enables parallel processing. Based on the DBP architecture, we formulate a weighted sum rate (WSR) maximization problem through joint beamforming and FA position design for FA-assisted multiuser multiple-input multiple-output (MU-MIMO) systems. To solve the WSR maximization problem, we propose a novel decentralized block coordinate ascent (BCA)-based algorithm that leverages matrix fractional programming (FP) and majorization-minimization (MM) methods. The proposed decentralized algorithm achieves low computational, communication, and storage costs, thus unleashing the potential of the DBP architecture. Simulation results show that our proposed algorithm under the DBP architecture reduces computational time by over 70% compared to centralized architectures with negligible WSR performance loss.

cs.IT↗

arXiv2025

FloE: On-the-Fly MoE Inference on Memory-constrained GPU

With the widespread adoption of Mixture-of-Experts (MoE) models, there is a growing demand for efficient inference on memory-constrained devices. While offloading expert parameters to CPU memory and loading activated experts on demand has emerged as a potential solution, the large size of activated experts overburdens the limited PCIe bandwidth, hindering the effectiveness in latency-sensitive scenarios. To mitigate this, we propose FloE, an on-the-fly MoE inference system on memory-constrained GPUs. FloE is built on the insight that there exists substantial untapped redundancy within sparsely activated experts. It employs various compression techniques on the expert's internal parameter matrices to reduce the data movement load, combined with low-cost sparse prediction, achieving perceptible inference acceleration in wall-clock time on resource-constrained devices. Empirically, FloE achieves a 9.3x compression of parameters per expert in Mixtral-8x7B; enables deployment on a GPU with only 11GB VRAM, reducing the memory footprint by up to 8.5x; and delivers a 48.7x inference speedup compared to DeepSpeed-MII on a single GeForce RTX 3090 - all with only a 4.4$\%$ - 7.6$\%$ average performance degradation.

cs.LG↗

arXiv2025

A Multi-Agent Reinforcement Learning Approach for Cooperative Air-Ground-Human Crowdsensing in Emergency Rescue

Mobile crowdsensing is evolving beyond traditional human-centric models by integrating heterogeneous entities like unmanned aerial vehicles (UAVs) and unmanned ground vehicles (UGVs). Optimizing task allocation among these diverse agents is critical, particularly in challenging emergency rescue scenarios characterized by complex environments, limited communication, and partial observability. This paper tackles the Heterogeneous-Entity Collaborative-Sensing Task Allocation (HECTA) problem specifically for emergency rescue, considering humans, UAVs, and UGVs. We introduce a novel ``Hard-Cooperative'' policy where UGVs prioritize recharging low-battery UAVs, alongside performing their sensing tasks. The primary objective is maximizing the task completion rate (TCR) under strict time constraints. We rigorously formulate this NP-hard problem as a decentralized partially observable Markov decision process (Dec-POMDP) to effectively handle sequential decision-making under uncertainty. To solve this, we propose HECTA4ER, a novel multi-agent reinforcement learning algorithm built upon a Centralized Training with Decentralized Execution architecture. HECTA4ER incorporates tailored designs, including specialized modules for complex feature extraction, utilization of action-observation history via hidden states, and a mixing network integrating global and local information, specifically addressing the challenges of partial observability. Furthermore, theoretical analysis confirms the algorithm's convergence properties. Extensive simulations demonstrate that HECTA4ER significantly outperforms baseline algorithms, achieving an average 18.42% increase in TCR. Crucially, a real-world case study validates the algorithm's effectiveness and robustness in dynamic sensing scenarios, highlighting its strong potential for practical application in emergency response.

cs.AI↗

arXiv2025

Adaptive Physics-Informed System Modeling with Control for Nonlinear Structural System Estimation

Accurately capturing the nonlinear dynamic behavior of structures remains a significant challenge in mechanics and engineering. Traditional physics-based models and data-driven approaches often struggle to simultaneously ensure model interpretability, noise robustness, and estimation optimality. To address this issue, this paper proposes an Adaptive Physics-Informed System Modeling with Control (APSMC) framework. By integrating Kalman filter-based state estimation with physics-constrained proximal gradient optimization, the framework adaptively updates time-varying state-space model parameters while processing real-time input-output data under white noise disturbances. Theoretically, this process is equivalent to real-time tracking of the Jacobian matrix of a nonlinear dynamical system. Within this framework, we leverage the theoretical foundation of stochastic subspace identification to demonstrate that, as observational data accumulates, the APSMC algorithm yields state-space model estimates that converge to the theoretically optimal solution. The effectiveness of the proposed framework is validated through numerical simulations of a Duffing oscillator and the seismic response of a frame structure, as well as experimental tests on a scaled bridge model. Experimental results show that, under noisy conditions, APSMC successfully predicts 19 consecutive 10-second time series using only a single initial 10-second segment for model updating, achieving a minimum normalized mean square error (NMSE) of 0.398%. These findings demonstrate that the APSMC framework not only offers superior online identification and denoising performance but also provides a reliable foundation for downstream applications such as structural health monitoring, real-time control, adaptive filtering, and system identification.

nlin.AO↗

arXiv2025

Accurate and Fast Channel Estimation for Fluid Antenna Systems with Diffusion Models

Fluid antenna systems (FAS) offer enhanced spatial diversity for next-generation wireless systems. However, acquiring accurate channel state information (CSI) remains challenging due to the large number of reconfigurable ports and the limited availability of radio-frequency (RF) chains -- particularly in high-dimensional FAS scenarios. To address this challenge, we propose an efficient posterior sampling-based channel estimator that leverages a diffusion model (DM) with a simplified U-Net architecture to capture the spatial correlation structure of two-dimensional FAS channels. The DM is initially trained offline in an unsupervised way and then applied online as a learned implicit prior to reconstruct CSI from partial observations via posterior sampling through a denoising diffusion restoration model (DDRM). To accelerate the online inference, we introduce a skipped sampling strategy that updates only a subset of latent variables during the sampling process, thereby reducing the computational cost with minimal accuracy degradation. Simulation results demonstrate that the proposed approach achieves significantly higher estimation accuracy and over 20x speedup compared to state-of-the-art compressed sensing-based methods, highlighting its potential for practical deployment in high-dimensional FAS.

cs.IT↗