Home Archive
Archive
2025 Volume 62 Issue 10  Published: 2025-10-01
    Research Highlights
  • Gerui Lü , Yuankang Zhao , Jiaxing Zhang , Heng Pan , Qinghua Wu , Zhenyu Li
    doi: 10.7544/issn1000-1239.202550536

    The Internet has become one of critical infrastructures in modern society, in which transmission performance directly determines the capability and value of the services it carries. With the rapid development of mobile and interactive Internet applications, low-latency transmission in large-scale Internet faces three technical challenges at the application layer, transport layer, and network layer, that are the mismatch between dynamic sending decisions of application data and available network resources, the incoordination between network transmission decisions and changes in network status, and the inconsistency between differentiated services and the processing logic of network devices. To address the bottlenecks of traditional hierarchical and single-point transmission optimization, we propose and construct a collaborative transmission technology system for large-scale Internet services. It introduces the technological achievements within this system in three aspects: intelligent transmission decision-making for applications, multi-dimensional collaborative transmission, and customized in-network transmission acceleration, including key insights, basic ideas and principles, and results of large-scale deployment. Finally, the next development trends of Internet transmission technology are prospected.

  • Special Issue on Data Security and Privacy Protection in New Distributed Networks and Systems
  • Zhecheng Li , Bo Zhang
    doi: 10.7544/issn1000-1239.202550402

    In the current context of cross-domain identity authentication and data sharing for industrial Internet of things (IIoT) devices, edge servers and blockchain technology play vital roles in enabling rapid data forwarding and secure storage. Existing mechanisms for cross-domain authentication and data sharing are facing a range of security risks, and are further challenged by high computational and communication overhead at the device level, as well as by overly complex authentication structures. To address these problems, we propose a lightweight cross-domain authentication and data sharing scheme based on blockchain technology, specifically designed for IIoT scenarios. The proposed scheme enables anonymous cross-domain authentication and secure data sharing for industrial devices. Furthermore, we introduce a distributed key generation mechanism to construct service keys. Based on service keys, a lightweight message authentication algorithm is designed to ensure the confidentiality of service-related information at the device side, while minimizing resource consumption to the greatest extent. A detailed security analysis demonstrates that the proposed scheme satisfies the critical security properties of unforgeability, confidentiality, unlinkability, and anonymity. In addition, performance evaluation results show that the proposed scheme achieves superior performance compared with existing solutions in terms of computational overhead, communication cost, and the efficiency of blockchain query operations.

  • Special Issue on Data Security and Privacy Protection in New Distributed Networks and Systems
  • Zhihong Deng , Chunming Tang , Taotao Li , Yaxuan He , Zhikang Zeng , Qi Chen
    doi: 10.7544/issn1000-1239.202550429

    With the rapid development of blockchain applications, light clients have become essential components in digital wallets and cross-chain communication due to their minimal storage and computational overhead. However, existing light client protocols face significant limitations in terms of storage and communication costs, security, and compatibility across different consensus mechanisms. We propose DecXClient, a general and secure light client protocol compatible with multiple consensus models. DecXClient introduces a dynamic validator set election mechanism based on chain quality attributes, eliminating reliance on trusted third parties. By integrating weighted multi-signature schemes, it enables a concise and efficient proof generation paradigm that enhances anti-fraud resilience while reducing resource consumption. Without depending on centralized relays, DecXClient maintains sublinear communication and storage complexity, effectively mitigating Eclipse attacks and other security threats. Experimental evaluations using real blockchain network environment demonstrate that the protocol produces state proofs as small as 1248 bytes—significantly outperforming existing schemes. The average proof generation latency ranges from 266.4 milliseconds to 2.5 seconds, with negligible verification time. These results validate the feasibility and practicality of DecXClient, offering a robust solution for lightweight blockchain access across diverse application scenarios.

  • Special Issue on Data Security and Privacy Protection in New Distributed Networks and Systems
  • Zhen Wu , Yang Zhao , Min Wang
    doi: 10.7544/issn1000-1239.202550382

    In the context of large-scale applications of distributed Internet of things (IoT), cryptographic technologies, as the underlying infrastructure for information security, are facing severe challenges from side-channel attacks (SCAs)—a physical-layer security threat. The SM4 block cipher algorithm, as China’s independently developed commercial cryptographic standard, has been deeply integrated into distributed IoT security protocols. However, its implementation-level vulnerabilities to side-channel attacks remain unresolved. Research on side-channel attacks targeting the SM4 key expansion algorithm is scarce, with existing methods primarily relying on multi-trace statistical characteristics, while single-trace attack approaches are under-explored. This paper proposes a single-trace side-channel attack method integrating Bayesian networks with side-channel attack modeling. By constructing a probabilistic graphical model and combining it with the belief propagation algorithm, the method efficiently infers round subkeys from a single power trace, ultimately recovering the master key. Simulation and practical experiments demonstrate the effectiveness of this approach, achieving an 85.74% success rate for master key recovery under ideal conditions. Even when substantial Gaussian white noise is added to practical power traces (resulting in a signal-to-noise ratio of only 10 dB), the success rate remains at 70%. Compared with traditional methods, this approach exhibits significant advantages in success rate, required power traces, and attack duration, offering new insights and technical tools for side-channel attack research on cryptographic devices in distributed IoT systems, as well as theoretical foundations for related defensive designs.

  • Special Issue on Data Security and Privacy Protection in New Distributed Networks and Systems
  • Yi Ding , Nengjian Lin , Yuntao Jiang , Yuhao Zhong , Mingsheng Cao
    doi: 10.7544/issn1000-1239.202550388

    As a common visual foundation model, the segment anything model (SAM) has been widely applied in various image segmentation tasks, but it exhibits vulnerability in the face of adversarial attacks. A adversarial transfer attack method FSGR (fragile section gradient robustness) has been proposed for vulnerable modules of SAM downstream models, which can effectively generate adversarial samples to attack SAM downstream models without understanding the details of fine-tuning in downstream. The core of FSGR lies in its ability to identify the most vulnerable parts and apply targeted attacks to these weak points. This method designs a strategy based on feature similarity to identify the most vulnerable module in SAM encoder and enhance the attack effectiveness in a targeted manner, and introduces gradient robust loss to eliminate the gradient difference between the target model and downstream task models. FSGR combines vulnerability layer attacks and gradient robust loss to achieve relative performance improvement on multiple datasets. The experimental results show that FSGR significantly reduces model performance in transfer attacks on various fine-tuning models, such as medical segmentation, shadow segmentation, and camouflage segmentation, demonstrating its correctness and practicality. Compared with baseline methods, FSGR not only performs well in attack success rate, but also achieves relative performance improvement by combining vulnerability layer attacks and gradient robust loss.

  • Special Issue on Data Security and Privacy Protection in New Distributed Networks and Systems
  • Caihong Yan , Xiaofeng Lu
    doi: 10.7544/issn1000-1239.202550442

    In the field of distributed artificial intelligence, federated learning models, like centrally trained models, are vulnerable to adversarial samples during reasoning. Currently, federated adversarial training has not been extensively explored and is facing two major challenges: 1) There is a trade-off between the model accuracy on clean samples and adversarial robustness, making it difficult to improve both simultaneously; 2) non-independent and identically distributed (Non-IID) data limit the performance improvement of federated adversarial training. To address these challenges, we propose the BTFAT framework, which breaks the trade-off between robustness and accuracy in federated adversarial training. The framework includes: 1) A decision-space tightening algorithm that performs initial intra-class localization using labels, while shrinking intra-class sample distances and increasing inter-class sample distances, thereby improving both robustness and accuracy; 2) A weight penalty optimization algorithm that uses the global model weights as the best unified target, penalizing local adversarial model training that deviates excessively, and assisting the decision-space algorithm to counter the impact of Non-IID data distributions. We theoretically analyze the key factors limiting the robustness and accuracy gains of adversarial training as well as the convergence of BTFAT. Simultaneously, we experimentally demonstrate that BTFAT comprehensively outperforms state-of-the-art baseline algorithms in terms of overall performance, convergence, time cost, and handling Non-IID data, providing a new perspective for research in federated adversarial training. Our code can be found in the website: https://anonymous.4open.science/r/BTFAT-11265.

  • Special Issue on Data Security and Privacy Protection in New Distributed Networks and Systems
  • Yingjie Zhou , Min Yang , Di Wu , Fanxing Liu , Wei Zhao , Yiran Deng , Jiancheng Lü
    doi: 10.7544/issn1000-1239.202550349

    In order to ensure the safety and efficient operation of cloud services, it is necessary to conduct comprehensive monitoring and analysis of various accessing behaviors and operations in cloud platforms, aiming at detecting potential abnormal behaviors in a timely and accurate manner. There are numerous anomaly detection tasks in cloud platforms. Existing approaches usually require task-specific customized design and fine-tuning, lacking generalizable model construction capabilities for various anomaly detection tasks. Moreover, this process heavily relies on machine learning expertise, making it challenging for domain experts to develop effective models suitable for practical applications. To address the above problems, an intelligent lightweight model construction solution is proposed for anomaly detection in cloud platforms. Compared with existing technology, the proposed solution 1) supports diverse types of anomaly detection tasks in cloud platforms, allowing domain experts to achieve a timely and self-service construction of task-oriented anomaly detection models by simply providing basic task-specific configuration information; 2) enables domain experts to intelligently construct lightweight models for target anomaly detection tasks through automated feature construction, feature optimization, and hyperparameter tuning, even with minimal machine learning and deep learning expertise. Both the case studies and the experimental results based on data collected from large-scale real-world cloud scenarios demonstrate the proposed solution could build models for relevant anomaly detection tasks in a timely and self-service manner, while strong detection capabilities are also achieved in the constructed models.

  • Special Issue on Data Security and Privacy Protection in New Distributed Networks and Systems
  • Feng Zhang , Hao Feng , Xiumin Zhou , Maochun Zhang , Zhenming Chen , Yuming Liu
    doi: 10.7544/issn1000-1239.202550399

    The critical challenge of detecting fabricated link attacks in low-latency agentic network environments, where conventional detection methods face significant limitations is investigated in this paper. The fundamental difficulty stems from three key factors: 1) the measured performance differences between normal and fabricated links often exhibit marginal variations that fall below meaningful detection thresholds; 2) dynamic background traffic patterns introduce substantial interference that obscures genuine attack signatures; and 3) coordinated overload attacks deliberately distort measurement accuracy. These factors collectively degrade the detection performance of existing methods by significantly reducing the observable distinction between legitimate and malicious links. To address these challenges, we propose a novel three-phase detection framework based on correlated link verification. The first phase implements an effective delay transformation method that reduces measurement errors while converting raw measurements into meaningful link performance metrics. The second phase employs a multipath transmission simulation technique that amplifies subtle performance disparities between correlated links into statistically observable features. The final phase establishes robust detection thresholds through the combined application of extreme value theory and probability distribution fitting methods. Comprehensive simulation results validate that our proposed method consistently outperforms existing approaches across diverse network scales and attack scenarios. The framework demonstrates particular effectiveness in low-latency agentic environments, maintaining superior detection performance while meeting stringent timing requirements. Specifically, the framework achieves a 20% to 30% improvement in F1 score compared with state-of-the-art methods. Notably, even under ultra-low latency conditions (e.g., 0.1ms link delay), the proposed method maintains a high detection rate of 95.23% with a false positive rate as low as 0.28%. These advancements provide a practical solution for securing next-generation autonomous networks against sophisticated link fabrication attacks.

  • Special Issue on Data Security and Privacy Protection in New Distributed Networks and Systems
  • Dongfen Li , Qiuyu Xiang , Zhikang Hu , Yuhang Yuan , Yuchen Sun , Han Xu , Zhengyu Song , Haomin Han , Dongxue Bai , Bin Duo , Rang Zhou
    doi: 10.7544/issn1000-1239.202550413

    Intrusion detection system (IDS) is a pivotal component in ensuring the security of network environments. However, conventional deep learning methodologies are characterised by elevated false alarm rates when confronted with infrequent attack features and challenges related to data imbalance. Additionally, these methodologies possess a limited capacity to effectively model high-dimensional sparse features. The ubiquity of the Internet of things (IoT) and edge devices has led to the emergence of intrusion detection data that exhibits distributed and heterogeneous trends. The centralised training of this data is hindered by privacy protection requirements, necessitating alternative approaches for data analysis and interpretation. In this regard, we put forward a federated classical hybrid quantum convolutional intrusion detection system (FedCQIDS), which involves the incorporation of a lightweight quantum convolutional neural network within the client-side local model. The convolutional layer is incorporated into the client-local model with the objective of enhancing resource utilisation through quantum bit multiplexing, augmenting the capacity to express low-frequency sparse attack features, and facilitating multi-client collaborative modelling and privacy protection by means of federated learning architecture. A series of related experiments are conducted on two distinct datasets, NSL-KDD and UNSW-NB15, and the findings indicate that FedCQIDS exhibits superior performance compared with conventional federated models with regard to accuracy and loss value. This is particularly evident in scenarios involving the identification of a limited number of classes of attacks, thereby underscoring its potential for application in the domain of privacy protection and efficient deployment.

  • Artificial Intelligence
  • Zheng Le , Yongting Hu , Yong Xu
    doi: 10.7544/issn1000-1239.202440207

    With the rapid advancement of artificial intelligence generation models and deepfakes, the techniques for generating talking face videos using various methods have become increasingly mature. Among them, audio-driven talking face video generation methods have attracted significant attention due to their remarkably realistic and natural output. Such methods utilize audio as a driving source to synthesize videos where the target character’s mouth movements synchronize with the audio, often combining image or video materials. Currently, these technologies are widely applied in fields such as virtual anchors, gaming animation, and film and television production, demonstrating vast prospects for development. However, the potential negative impacts of this technology are also becoming apparent. Improper or abusive use could lead to serious political and economic consequences. In this context, research on identifying various types of facial forgery videos has emerged. Our research primarily assesses the authenticity of videos by detecting the veracity of individual video frames or the spatio-temporal consistency of video sequences. Firstly, we systematically analyze the classic algorithms and latest advancements in audio-driven talking face video generation tasks based on the timeline and the development history of foundational models. Secondly, we exhaustively list the commonly used datasets and evaluation criteria for this task, conducting comprehensive comparisons across multiple dimensions. Subsequently, we meticulously analyze and summarize the forgery facial video identification task, categorizing it based on whether the discrimination technology focuses on individual video frames or multiple frames, and also summarize its commonly used datasets and evaluation criteria. Finally, we outline the challenges and future directions in this research field, aiming to provide valuable references and support for subsequent related research.

  • Artificial Intelligence
  • Silong Dai , Pengfei Wang , Xiaoling Wang
    doi: 10.7544/issn1000-1239.202440191

    In recent years, the rapid advancement of deep learning technology has introduced innovative solutions to the field of face de-identification. Compared with traditional image processing techniques, deep generative models have demonstrated significant advantages in this domain, including high-quality image generation and robust model performance. This article reviews and synthesizes the theoretical explorations and research outcomes of deep learning technology in addressing face de-identification challenges. It begins by outlining the network architectures and fundamental principles employed in deep learning for face de-identification. It then delves into the de-identification methods based on these technologies, covering key techniques such as face swapping and feature perturbation, and introduces the standard experimental metrics used to evaluate these methods. Furthermore, the article summarizes the main challenges currently faced by the technology, such as the stability of posture and expression, attribute disentanglement, and the adaptability to video applications, and looks forward to the pressing issues that future research needs to address. Ultimately, this article emphasizes the importance of deep learning technology in the field of face de-identification and points out the direction for future research. It aims to provide readers with in-depth insights into the field of face de-identification and inspire new ideas and directions for future studies.

  • Artificial Intelligence
  • Rui Wang , Liuyang Zhang , Zhiyong Gao , Tongyun Jiang
    doi: 10.7544/issn1000-1239.202440385

    With the rapid development of large-scale model technology, these models have exhibited remarkable performance in fields such as natural language processing and computer vision, becoming essential tools for addressing complex issues and drawing significant interest from both the scientific community and the industry. Nonetheless, current cloud-platform-based schemes for training and inference of large models face multiple challenges, including high expenses, restricted scalability, and information security risks. As the scale of model parameters expands continually, the need for low-cost, efficient training and inference methods grows ever more pressing. Carrying out collaborative training and inference of large models on edge devices can dramatically decrease latency and bandwidth demands, concurrently reinforcing data privacy and operational efficiency. This strategy furnishes vital technological support for the economical deployment of large models across a variety of contexts, thereby evolving into one of the prominent research hotspots. This article conducts a thorough investigation of research pertinent to large models in the context of edge intelligence, with an in-depth analysis and discourse primarily focused on two aspects: edge-based training and inference of large models. Ultimately, it outlines the challenges in the progression of large model technologies tailored for edge intelligence and delineates future prospects. The ambition is to stimulate a heightened comprehension and intensified attention from both academic and industrial sectors towards technologies involving large models for edge intelligence, thereby encouraging further scholarly exploration in this thriving domain.

  • Artificial Intelligence
  • Xinyue Liu , Hailian Yin , Yalei Zang , Wenlong Wu , Junnan Zhuo , Fengru Xu , lüying Chen , Weihua Ma , Bohan Li
    doi: 10.7544/issn1000-1239.202220749

    Sequential recommendation systems aim to predict users’ next actions based on the preferences learned from their historical behaviors. There are still fundamental challenges for sequential recommender. First, with the popularization of online services, recommender need serve both the warm-start and cold-start users. However, the most existing models depending on user-item interactions lose merits due to the difficulty of learning sequential dependencies with limited interactions. Second, users’ behaviors in their historical sequences are often implicit and complex due to the objective variability of reality and the subjective randomness of users’ intentions. It is difficult to capture the dynamic transition patterns from these user-item interactions. In this work, we propose a graph-based interpolation sequential recommender with deformable convolutional network (GISDCN). For cold-start users, we reconstruct item sequences into a graph to infer users’ possible preferences. To capture the complex sequential dependencies, we employ the deformable convolutional network to generate more robust and flexible filters. Finally, we conduct comprehensive experiments and verify the effectiveness of our model. The experimental results demonstrate that GISDCN outperforms most of the state-of-the-art models at cold-start conditions.

  • Artificial Intelligence
  • Jia Zhao , Qiumin Hu , Renbin Xiao , Jeng-Shyang Pan , Li Lü , Hui Wang , Tanghuai Fan
    doi: 10.7544/issn1000-1239.202440019

    Aiming at the problems of Pareto dominance failure, weak optimization ability, and slow convergence speed of multi-objective firefly algorithm in solving many-objective optimization problems, we propose a many-objective firefly algorithm based on reference point guidance and multiple cooperative strategies (MaOFA-RR). This algorithm presets a set of uniformly distributed reference points in the objective space. By examining the distance relationship between fireflies and reference points, it distinguishes between guide fireflies and ordinary fireflies to replace Pareto dominance, thereby increasing selection pressure. Three evolutionary strategies are employed to update the positions of the fireflies: guide fireflies explore the local space, while ordinary fireflies learn from guide fireflies or explore the global space based on a distance threshold, enhancing the algorithm’s optimization capability and convergence speed. Finally, the algorithm ingeniously integrates the idea of opposition-based learning. In the process of opposition-based learning, for each firefly in the population, its opposite position is calculated in the solution space. By adding these opposite solutions to the original population, the algorithm effectively expands the scope of population search, significantly improving the possibility of discovering better solutions. We conduct comprehensive experiments by comparing MaOFA-RR with eight recent many-objective evolutionary algorithms. The experimental results show that MaOFA-RR exhibits efficient performance in handing many-objective optimization problems.

  • Artificial Intelligence
  • Hao Yao , Jinhui Xiong , Chunsheng Li , Changxing Wu
    doi: 10.7544/issn1000-1239.202440511

    Implicit discourse relation recognition aims at automatically identifying semantic relations (such as Comparison) between two arguments (sentence or clause) in the absence of explicit connectives. Existing methods have confirmed that the introduction of phrase information can effectively boost the performance. However, there are still the following shortcomings: 1) These models typically rely on syntactic parsers and do not fully capture the interactions among words, phrases, and arguments. 2) The problem of data sparsity often occurs during training when incorporating the phrase information. To address the above issues, we propose an implicit discourse relation recognition model based on multi-granularity information interaction (MGII) and develop a chain decoding-inspired data augmentation method (DAM). Specifically, our proposed model is designed to automatically acquire semantic representations of n-grams using a stacked convolutional neural network. It then explicitly models the interactions among words, phrases and arguments based on Transformer layers and ultimately predicts multi-level discourse relationships in a chain-decoding way. DAM proposed in our paper simultaneously pretrains both the encoding and decoding modules, enabling the effective utilization of massive explicit discourse data, which are naturally annotated by connectives, to mitigate the issue of data sparsity. The proposed method significantly outperforms recent benchmark models on PDTB datasets. Furthermore, it does not rely on syntactic parsers, demonstrating strong applicability.

  • Software Technology
  • Qingxiao Sun , Hailong Yang
    doi: 10.7544/issn1000-1239.202440612

    Stencil computations are widely adopted in scientific applications. Many HPC platforms utilize the high computation capability of GPUs to accelerate Stencil computations. In recent years, Stencils have become more complex in terms of Stencil order, memory accesses, and computation patterns. To adapt Stencil computations to GPU architectures, the academic community has proposed a variety of optimization techniques based on streaming and tiling. Due to the diversity of Stencil computational patterns and GPU architectures, no single optimization technique fits all Stencil instances. Therefore, researchers have proposed Stencil auto-tuning mechanisms to conduct parameter searches for a given combination of optimization techniques. However, existing mechanisms introduce huge offline profiling costs and online prediction overhead, unable to be flexible to arbitrary Stencil patterns. To address the above problems, we propose a generalized Stencil auto-tuning framework GeST, which achieves the ultimate performance optimization of Stencil computations on GPU platforms. Specifically, GeST constructs the global search space through the zero-padding format, quantifying parameter correlations via the coefficient of variation to generate parameter groups. After that, GeST iteratively selects parameter values from the parameter groups, adjusting the sampling ratio according to the reward policy and avoiding redundant execution through Hash coding. The experimental results show that GeST can identify better-performing parameter settings in a short time compared with other state-of-the-art auto-tuning work.

  • Software Technology
  • Xiaoli Zhang , Gang Chen , Yingying Ma , Ping Hu
    doi: 10.7544/issn1000-1239.202440672

    As the scale of AI grows rapidly, errors in deep learning applications are also increasing. Existing popular deep learning frameworks are mostly built on the dynamical-typed language Python, which lacks type checking mechanisms. This leads to many errors that cannot be eliminated through type checking at the compilation stage. We propose a strongly-typed functional programming style deep learning framework based on the theorem prover Coq. The framework features typed tensor structures and a powerful static type checking capability. The experimental results demonstrate that the framework can automatically, quickly, and effectively detect shape mismatch errors in deep learning models, and it has greater advantages in terms of speed and detection capability compared with other checking tools. Furthermore, we design and implement a set of rewriting rules that translate functional programming models into C code, realizing the translation from functional neural network operator expressions to multi-core parallel OpenMP C code. According to the results of multiple sets of experiments, the C code for operators generated by this method is on par with manually written code. Furthermore, the speed of the generated neural network operator C code with multi-core parallel optimization is improved by 4−10 times compared with the sequentially executed operator C code. Additionally, the generated C operators are highly secure and can effectively avoid common issues such as out-of-bounds indexing and memory allocation errors in manually written code.