arXiv subjects

Yuxiang Zhang

Publications and source records attributed to Yuxiang Zhang.

At least 91 records · Page 5Linked to original sources

arXiv2025

ClassEval-T: Evaluating Large Language Models in Class-Level Code Translation

In recent years, Large Language Models (LLMs) have dramatically advanced the performance of automated code translation, making their computational accuracy score reach up to over 80% on many previous benchmarks. However, most code samples in these benchmarks are short, standalone, statement/method-level, and algorithmic, which is not aligned with practical coding tasks. Therefore, it is still unknown the actual capability of LLMs in translating code samples written for daily development. To achieve this, we construct a class-level code translation benchmark, ClassEval-T, and make the first attempt to extensively assess recent LLMs' performance on class-level code translation. ClassEval-T is extended from ClassEval, a well-known class-level Python code generation benchmark consisting of multiple practical coding topics, such as database operation and game design, and diverse contextual dependencies (e.g., fields, methods, and libraries). It cost us 360 person-hours to accomplish the manual migration to Java and C++ with complete code samples and associated test suites. Subsequently, we design three translation strategies (i.e., holistic, min-dependency, and standalone) for class-level code translations and evaluate eight recent LLMs of commercial, general, and code kinds in diverse families and sizes on ClassEval-T. Experimental results demonstrate a remarkable performance drop compared with the most widely studied method-level code translation benchmark, and obvious discrepancies among LLMs appear, showing the effectiveness of ClassEval-T in measuring recent LLMs. Afterwards, we further discuss the usage scenarios for diverse translation strategies and LLMs' ability to dependency awareness when translating class samples. Finally, 1,243 failure cases made by the best-performing LLM under test are analyzed and categorized in this paper for practical guidance and future enlightenment.

cs.SE↗

arXiv2025

Research and Experimental Validation for 3GPP ISAC Channel Modeling Standardization

Integrated Sensing and Communication (ISAC) is considered a key technology in 6G networks. An accurate sensing channel model is crucial for the design and sensing performance evaluation of ISAC systems. The widely used Geometry-Based Stochastic Model (GBSM), typically applied in standardized channel modeling, mainly focuses on the statistical fading characteristics of the channel. However, it fails to capture the characteristics of targets in ISAC systems, such as their positions and velocities, as well as the impact of the targets on the background. To address this issue, this paper proposes an extended GBSM (E-GBSM) sensing channel model that incorporates newly discovered channel characteristics into a unified modeling framework. In this framework, the sensing channel is divided into target and background channels. For the target channel, the model introduces a concatenated modeling approach, while for the background channel, a parameter called the power control factor is introduced to assess impact of the target on the background channel, making the modeling framework applicable to both mono-static and bi-static sensing modes. To validate the proposed model's effectiveness, measurements of target and background channels are conducted in both indoor and outdoor scenarios, covering various sensing targets such as metal plates, reconfigurable intelligent surfaces, human bodies, UAVs, and vehicles. The experimental results provide important theoretical support and empirical data for the standardization of ISAC channel modeling.

eess.SP↗

arXiv2025

A Novel Environment Object Modeling Method for Vehicular ISAC Scenarios

Integrated Sensing and Communication (ISAC), as a fundamental technology of 6G, empowers Vehicle-to-Everything (V2X) systems with enhanced sensing capabilities. One of its promising applications is the reliance on constructed maps for vehicle positioning. Traditional positioning methods primarily rely on Line-of-Sight (LOS), but in urban vehicular scenarios, obstructions often result in predominantly Non-Line-of-Sight (NLOS) conditions. Existing research indicates that NLOS paths, characterized by one-bounce reflection on building walls with determined delay and angle, can support sensing and positioning. However, experimental validation remains insufficient. To address this gap, channel measurements are conducted in an urban street to explore the existence of strong reflected paths in the presence of a vehicle target. The results show significant power contribution from NLOS paths, with large Environmental Objects (EOs) playing a key role in shaping NLOS propagation. Then, a novel model for EO reflection is proposed to extend the Geometry-Based Stochastic Model (GBSM) for ISAC channel standardization. Simulation results validate the model's ability to capture EO's power and position characteristics, showing that higher EO-reflected power and closer distance to Rx reduce Delay Spread (DS), which is more favorable for positioning. This model provides theoretical guidance and empirical support for ISAC positioning algorithms and system design in vehicular scenarios.

eess.SP↗

arXiv2025

Agent models: Internalizing Chain-of-Action Generation into Reasoning models

Traditional agentic workflows rely on external prompts to manage interactions with tools and the environment, which limits the autonomy of reasoning models. We position \emph{Large Agent Models (LAMs)} that internalize the generation of \emph{Chain-of-Action (CoA)}, enabling the model to autonomously decide when and how to use external tools. Our proposed AutoCoA framework combines supervised fine-tuning (SFT) and reinforcement learning (RL), allowing the model to seamlessly switch between reasoning and action while efficiently managing environment interactions. Main components include step-level action triggering, trajectory-level CoA optimization, and an internal world model to reduce real-environment interaction costs. Evaluations on open-domain QA tasks demonstrate that AutoCoA-trained agent models significantly outperform ReAct-based workflows in task completion, especially in tasks that require long-term reasoning and multi-step actions. Code and dataset are available at https://github.com/ADaM-BJTU/AutoCoA

cs.AI↗

arXiv2025

LoR2C : Low-Rank Residual Connection Adaptation for Parameter-Efficient Fine-Tuning

In recent years, pretrained large language models have demonstrated outstanding performance across various natural language processing tasks. However, full-parameter fine-tuning methods require adjusting all model parameters, leading to immense computational resource demands. Although parameter-efficient fine-tuning methods like LoRA have significantly reduced the number of parameters, they still face challenges such as gradient vanishing and the potential for further parameter reduction. To address these issues, this paper proposes a novel parameter-efficient fine-tuning method called LoR2C (Low-Rank Residual Connection Adaptation). LoR2C introduces residual connections with low-rank matrices within the model layers, which not only reduces the number of fine-tuning parameters but also effectively alleviates the gradient vanishing problem. Additionally, this paper presents three optimization variants of LoR2C: ShareLoR2C, MergeLoR2C, and InjectLoR2C. These variants further improve parameter efficiency and model performance through parameter sharing, module merging, and injection mechanisms, respectively. Experimental results on multiple natural language understanding and natural language generation tasks demonstrate that LoR2C and its optimized variants significantly reduce parameter overhead while maintaining or even improving performance, outperforming existing mainstream parameter-efficient fine-tuning methods.Our code is publicly available at https://github.com/Oblivioniss/LoR2C.

cs.CL↗

arXiv2025

ChartMimic: Evaluating LMM's Cross-Modal Reasoning Capability via Chart-to-Code Generation

We introduce a new benchmark, ChartMimic, aimed at assessing the visually-grounded code generation capabilities of large multimodal models (LMMs). ChartMimic utilizes information-intensive visual charts and textual instructions as inputs, requiring LMMs to generate the corresponding code for chart rendering. ChartMimic includes 4,800 human-curated (figure, instruction, code) triplets, which represent the authentic chart use cases found in scientific papers across various domains (e.g., Physics, Computer Science, Economics, etc). These charts span 18 regular types and 4 advanced types, diversifying into 201 subcategories. Furthermore, we propose multi-level evaluation metrics to provide an automatic and thorough assessment of the output code and the rendered charts. Unlike existing code generation benchmarks, ChartMimic places emphasis on evaluating LMMs' capacity to harmonize a blend of cognitive capabilities, encompassing visual understanding, code generation, and cross-modal reasoning. The evaluation of $3$ proprietary models and 14 open-weight models highlights the substantial challenges posed by ChartMimic. Even the advanced GPT-4o, InternVL2-Llama3-76B only achieved an average score across Direct Mimic and Customized Mimic tasks of 82.2 and 61.6, respectively, indicating significant room for improvement. We anticipate that ChartMimic will inspire the development of LMMs, advancing the pursuit of artificial general intelligence.

cs.SE↗

arXiv2025

Gaussian Difference: Find Any Change Instance in 3D Scenes

Instance-level change detection in 3D scenes presents significant challenges, particularly in uncontrolled environments lacking labeled image pairs, consistent camera poses, or uniform lighting conditions. This paper addresses these challenges by introducing a novel approach for detecting changes in real-world scenarios. Our method leverages 4D Gaussians to embed multiple images into Gaussian distributions, enabling the rendering of two coherent image sequences. We segment each image and assign unique identifiers to instances, facilitating efficient change detection through ID comparison. Additionally, we utilize change maps and classification encodings to categorize 4D Gaussians as changed or unchanged, allowing for the rendering of comprehensive change maps from any viewpoint. Extensive experiments across various instance-level change detection datasets demonstrate that our method significantly outperforms state-of-the-art approaches like C-NERF and CYWS-3D, especially in scenarios with substantial lighting variations. Our approach offers improved detection accuracy, robustness to lighting changes, and efficient processing times, advancing the field of 3D change detection.

cs.CV↗

arXiv2025

A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models

The rise of foundation models has transformed machine learning research, prompting efforts to uncover their inner workings and develop more efficient and reliable applications for better control. While significant progress has been made in interpreting Large Language Models (LLMs), multimodal foundation models (MMFMs) - such as contrastive vision-language models, generative vision-language models, and text-to-image models - pose unique interpretability challenges beyond unimodal frameworks. Despite initial studies, a substantial gap remains between the interpretability of LLMs and MMFMs. This survey explores two key aspects: (1) the adaptation of LLM interpretability methods to multimodal models and (2) understanding the mechanistic differences between unimodal language models and crossmodal systems. By systematically reviewing current MMFM analysis techniques, we propose a structured taxonomy of interpretability methods, compare insights across unimodal and multimodal architectures, and highlight critical research gaps.

cs.LG↗

arXiv2025

Road to 6G Digital Twin Networks: Multi-Task Adaptive Ray-Tracing as a Key Enabler

As a virtual, synchronized replica of physical network, the digital twin network (DTN) is envisioned to sense, predict, optimize and manage the intricate wireless technologies and architectures brought by 6G. Given that the properties of wireless channel fundamentally determine the system performances from the physical layer to network layer, it is a critical prerequisite that the invisible wireless channel in physical world be accurately and efficiently twinned. To support 6G DTN, this paper first proposes a multi-task adaptive ray-tracing platform for 6G (MART-6G) to generate the channel with 6G features, specially designed for DTN online real-time and offline high-accurate tasks. Specifically, the MART-6G platform comprises three core modules, i.e., environment twin module to enhance the sensing ability of dynamic environment; RT engine module to incorporate the main algorithms of propagations, accelerations, calibrations, 6G-specific new features; and channel twin module to generate channel multipath, parameters, statistical distributions, and corresponding three-dimensional (3D) environment information. Moreover, MART-6G is tailored for DTN tasks through the adaptive selection of proper sensing methods, antenna and material libraries, propagation models and calibration strategy, etc. To validate MART-6G performance, we present two real-world case studies to demonstrate the accuracy, efficiency and generality in both offline coverage prediction and online real-time channel prediction. Finally, some open issues and challenges are outlined to further support future diverse DTN tasks.

eess.SP↗

arXiv2025

Step-Audio: Unified Understanding and Generation in Intelligent Speech Interaction

Real-time speech interaction, serving as a fundamental interface for human-machine collaboration, holds immense potential. However, current open-source models face limitations such as high costs in voice data collection, weakness in dynamic control, and limited intelligence. To address these challenges, this paper introduces Step-Audio, the first production-ready open-source solution. Key contributions include: 1) a 130B-parameter unified speech-text multi-modal model that achieves unified understanding and generation, with the Step-Audio-Chat version open-sourced; 2) a generative speech data engine that establishes an affordable voice cloning framework and produces the open-sourced lightweight Step-Audio-TTS-3B model through distillation; 3) an instruction-driven fine control system enabling dynamic adjustments across dialects, emotions, singing, and RAP; 4) an enhanced cognitive architecture augmented with tool calling and role-playing abilities to manage complex tasks effectively. Based on our new StepEval-Audio-360 evaluation benchmark, Step-Audio achieves state-of-the-art performance in human evaluations, especially in terms of instruction following. On open-source benchmarks like LLaMA Question, shows 9.3% average performance improvement, demonstrating our commitment to advancing the development of open-source multi-modal language technologies. Our code and models are available at https://github.com/stepfun-ai/Step-Audio.

cs.CL↗

arXiv2025

SpeechAct: Towards Generating Whole-body Motion from Speech

This paper addresses the problem of generating whole-body motion from speech. Despite great successes, prior methods still struggle to produce reasonable and diverse whole-body motions from speech. This is due to their reliance on suboptimal representations and a lack of strategies for generating diverse results. To address these challenges, we present a novel hybrid point representation to achieve accurate and continuous motion generation, e.g., avoiding foot skating, and this representation can be transformed into an easy-to-use representation, i.e., SMPL-X body mesh, for many applications. To generate whole-body motion from speech, for facial motion, closely tied to the audio signal, we introduce an encoder-decoder architecture to achieve deterministic outcomes. However, for the body and hands, which have weaker connections to the audio signal, we aim to generate diverse yet reasonable motions. To boost diversity in motion generation, we propose a contrastive motion learning method to encourage the model to produce more distinctive representations. Specifically, we design a robust VQ-VAE to learn a quantized motion codebook using our hybrid representation. Then, we regress the motion representation from the audio signal by a translation model employing our contrastive motion learning method. Experimental results validate the superior performance and the correctness of our model. The project page is available for research purposes at http://cic.tju.edu.cn/faculty/likun/projects/SpeechAct.

cs.CV↗

arXiv2025

Don't Command, Cultivate: An Exploratory Study of System-2 Alignment

The o1 system card identifies the o1 models as the most robust within OpenAI, with their defining characteristic being the progression from rapid, intuitive thinking to slower, more deliberate reasoning. This observation motivated us to investigate the influence of System-2 thinking patterns on model safety. In our preliminary research, we conducted safety evaluations of the o1 model, including complex jailbreak attack scenarios using adversarial natural language prompts and mathematical encoding prompts. Our findings indicate that the o1 model demonstrates relatively improved safety performance; however, it still exhibits vulnerabilities, particularly against jailbreak attacks employing mathematical encoding. Through detailed case analysis, we identified specific patterns in the o1 model's responses. We also explored the alignment of System-2 safety in open-source models using prompt engineering and supervised fine-tuning techniques. Experimental results show that some simple methods to encourage the model to carefully scrutinize user requests are beneficial for model safety. Additionally, we proposed a implementation plan for process supervision to enhance safety alignment. The implementation details and experimental results will be provided in future versions.

cs.CL↗

arXiv2024

An Approximate Wave-Number Domain Expression for Near-Field XL-array Channel

As Extremely large-scale array (XL-array) technology advances and carrier frequency rises, the near-field effects in communication are intensifying. In near-field conditions, channels exhibit a diffusion phenomenon in the angular domain, existing research indicates that this phenomenon can be leveraged for efficient parameter estimation and beam training. However, the channel model in angular domain lacks closed-form analysis, making the time complexity of the corresponding algorithm high. To address this issue, this paper analyzes the near-field diffusion effect in the wave-number domain, where the wave-number domain can be viewed as the continuous form of the angular domain. A closed-form approximate wave-number domain expression is proposed, based on the Principle of Stationary Phase. Subsequently, we derive a simplified expression for the case where the user distance is much larger than the array aperture, which is more concise. Subsequently, we verify the accuracy of the proposed approximate expression through simulations and demonstrate its effectiveness using a beam training example. Results indicate that the beam training scheme, improved by the wave-number domain approximation model, can effectively estimate near-field user parameters and perform beam training using far-field DFT codebooks. Moreover, its performance surpasses that of existing DFT codebook-based beam training methods.

eess.SP↗

arXiv2024

COMO: Cross-Mamba Interaction and Offset-Guided Fusion for Multimodal Object Detection

Single-modal object detection tasks often experience performance degradation when encountering diverse scenarios. In contrast, multimodal object detection tasks can offer more comprehensive information about object features by integrating data from various modalities. Current multimodal object detection methods generally use various fusion techniques, including conventional neural networks and transformer-based models, to implement feature fusion strategies and achieve complementary information. However, since multimodal images are captured by different sensors, there are often misalignments between them, making direct matching challenging. This misalignment hinders the ability to establish strong correlations for the same object across different modalities. In this paper, we propose a novel approach called the CrOss-Mamba interaction and Offset-guided fusion (COMO) framework for multimodal object detection tasks. The COMO framework employs the cross-mamba technique to formulate feature interaction equations, enabling multimodal serialized state computation. This results in interactive fusion outputs while reducing computational overhead and improving efficiency. Additionally, COMO leverages high-level features, which are less affected by misalignment, to facilitate interaction and transfer complementary information between modalities, addressing the positional offset challenges caused by variations in camera angles and capture times. Furthermore, COMO incorporates a global and local scanning mechanism in the cross-mamba module to capture features with local correlation, particularly in remote sensing images. To preserve low-level features, the offset-guided fusion mechanism ensures effective multiscale feature utilization, allowing the construction of a multiscale fusion data cube that enhances detection performance.

cs.CV↗

arXiv2024

OpenRFT: Adapting Reasoning Foundation Model for Domain-specific Tasks with Reinforcement Fine-Tuning

OpenAI's recent introduction of Reinforcement Fine-Tuning (RFT) showcases the potential of reasoning foundation model and offers a new paradigm for fine-tuning beyond simple pattern imitation. This technical report presents \emph{OpenRFT}, our attempt to fine-tune generalist reasoning models for domain-specific tasks under the same settings as RFT. OpenRFT addresses two key challenges of lacking reasoning step data and the limited quantity of training samples, by leveraging the domain-specific samples in three ways: question augmentation, synthesizing reasoning-process data, and few-shot ICL. The evaluation is conducted on SciKnowEval, where OpenRFT achieves notable performance gains with only $100$ domain-specific samples for each task. More experimental results will be updated continuously in later versions. Source codes, datasets, and models are disclosed at: https://github.com/ADaM-BJTU/OpenRFT

cs.AI↗

arXiv2024

InterDance:Reactive 3D Dance Generation with Realistic Duet Interactions

Humans perform a variety of interactive motions, among which duet dance is one of the most challenging interactions. However, in terms of human motion generative models, existing works are still unable to generate high-quality interactive motions, especially in the field of duet dance. On the one hand, it is due to the lack of large-scale high-quality datasets. On the other hand, it arises from the incomplete representation of interactive motion and the lack of fine-grained optimization of interactions. To address these challenges, we propose, InterDance, a large-scale duet dance dataset that significantly enhances motion quality, data scale, and the variety of dance genres. Built upon this dataset, we propose a new motion representation that can accurately and comprehensively describe interactive motion. We further introduce a diffusion-based framework with an interaction refinement guidance strategy to optimize the realism of interactions progressively. Extensive experiments demonstrate the effectiveness of our dataset and algorithm.

cs.CV↗

arXiv2024

Wireless Environmental Information Theory: A New Paradigm towards 6G Online and Proactive Environment Intelligence Communication

The channel is one of the five critical components of a communication system, and its ergodic capacity is based on all realizations of statistic channel model. This statistical paradigm has successfully guided the design of mobile communication systems from 1G to 5G. However, this approach relies on offline channel measurements in specific environments, and the system passively adapts to new environments, resulting in deviation from the optimal performance. With the pursuit of higher capacity and data rate of 6G, especially facing the ubiquitous environments, there is an urgent need for a new paradigm to combat the randomness of channel, i.e., more proactive and online manner. Motivated by this, we propose an environment intelligence communication (EIC) based on wireless environmental information theory (WEIT) for 6G. The proposed EIC architecture is composed of three steps: Firstly, wireless environmental information (WEI) is acquired using sensing techniques. Then, leveraging WEI and channel data, AI techniques are employed to predict channel fading, thereby mitigating channel uncertainty. Thirdly, the communication system autonomously determines the optimal air-interface transmission strategy based on real-time channel predictions, enabling intelligent interaction with the physical environment. To make this attractive paradigm shift from theory to practice, we answer three key problems to establish WEIT for the first time. How should WEI be defined? Can it be quantified? Does it hold the same properties as statistical communication information? Furthermore, EIC aided by WEI (EIC-WEI) is validated across multiple air-interface tasks, including CSI prediction, beam prediction, and radio resource management. Simulation results demonstrate that the proposed EIC-WEI significantly outperforms the statistical paradigm in decreasing overhead and performance optimization.

cs.IT↗

arXiv2024

o1-Coder: an o1 Replication for Coding

The technical report introduces O1-CODER, an attempt to replicate OpenAI's o1 model with a focus on coding tasks. It integrates reinforcement learning (RL) and Monte Carlo Tree Search (MCTS) to enhance the model's System-2 thinking capabilities. The framework includes training a Test Case Generator (TCG) for standardized code testing, using MCTS to generate code data with reasoning processes, and iteratively fine-tuning the policy model to initially produce pseudocode and then generate the full code. The report also addresses the opportunities and challenges in deploying o1-like models in real-world applications, suggesting transitioning to the System-2 paradigm and highlighting the imperative for world model construction. Updated model progress and experimental results will be reported in subsequent versions. All source code, curated datasets, as well as the derived models are disclosed at https://github.com/ADaM-BJTU/O1-CODER .

cs.SE↗