Home Archive
Archive
2026 Volume 63 Issue 1  Published: 2026-01-01
    Computer Architecture
  • Shuo Feng , Dongdong Lu , Fei Yin , Jianxin Yang , Dongsong Ban , Jun He , Shiyun Yan , Yuan Li , Haoyu Ju
    doi: 10.7544/issn1000-1239.202440675

    As one of the most distinctive representations for distributed computing, an end-edge-cloud collaborative computing system is capable of effectively bringing such applications as Internet of things, large-scale language models, and digital twins into real life. By mimicking the neural activities that take place in the human brain, brain-inspired computing has various advantages, including energy efficiency, high speed, high error-tolerance, and desirable scalability. Through leveraging the event-driven mechanism and the sparsity in spike generation of spiking neural networks, the real-time processing capability and energy efficiency of an end-edge-cloud collaborative computing system can be significantly improved. In this paper, a distributed computing-oriented ISA (instruction set architecture) design for brain-inspired intelligence CPU is studied. Bearing in mind the requirements of delay-sensitive, low-power, and high diversity for end devices, we focus on the software-hardware interface, i.e., ISA, and propose a hardware design that is rooted in the current systems, easy to implement and upgrade, safety-aware and self-controllable, and compatible with heterogeneous architectures. Along with a corresponding CPU micro-architecture design, a dozen of instructions specifically conceived for brain-inspired computing are proposed based on a well established ISA, which sets the stage for empowering distributed computing systems with brain-inspired computing. We firmly believe that this paper has discovered a fruitful research field, and sincerely hope that more interests can be stimulated in ISA designs for brain-inspired intelligence CPU in both academic and industrial communities.

  • Computer Architecture
  • Mingjian He , Hua Wang
    doi: 10.7544/issn1000-1239.202550001

    Multithread programming models are widely employed to facilitate application execution on multiple cores. However, interference caused by competition for the last level cache (LLC) among concurrently executed threads can lead to performance degradation. Intel cache allocation technology (CAT) provides a mechanism to assign cache ways to cores for achieving cache isolation. Unfortunately, prior studies on CAT-based partitioning are not applicable to multithreaded applications due to two reasons. Firstly, they are tailored for multi-application scenarios rather than single-application scenarios involving multiple related threads. Secondly, they are designed for improving IPC (instructions per cycle), which is not a suitable performance indicator in multithreaded scenarios. To bridge this gap, we present LPart, an intelligent cache partitioning technique for multithreaded applications that significantly improve throughput in multicores with way-partitioned caches. LPart leverages deep reinforcement learning to assign an appropriate number of LLC ways for different threads within an application, without needing any prior knowledge of the application characteristics. We evaluate LPart on a microbenchmark, Redis, a commercial production-ready distributed storage system, and multiple application scenarios. The experimental results indicate that, compared with the default configuration, LPart achieves speedups up to 26.9%, 8.1%, 9.8%, and 24.1% respectively without requiring any changes to the operating system.

  • Computer Architecture
  • Hongliang Wang , Zhenhua Guo , Qi Mu , Shengcai Lu , Yaming Xu , Quanquan Yu
    doi: 10.7544/issn1000-1239.202440833

    In recent years, although Chinese domestic field programmable gate array (FPGA) manufacturers have developed rapidly, they still face challenges when deploying FPGA heterogeneous accelerators in data centers. Compared with international manufacturers such as Xilinx (now AMD) and Intel, domestic manufacturers generally lack solutions for high-speed transmission between PCIe devices and hosts, especially in the field of high-performance direct memory access (DMA) controller design, where there are obvious shortcomings. To solve this problem, we design and implement a PCIe-based multi-channel chained DMA controller. By using an independent descriptor controller to manage each channel, sharing data movers, and reducing the consumption of FPGA logic resources, this design improves resource efficiency. The adoption of a chain structure for descriptor management reduces CPU interrupt pressure while meeting the requirements for continuous high-speed transmission between hosts and devices. An innovative architecture for asynchronous internal information pre-processing is developed, enabling data stream processing that significantly improves bandwidth utilization and transmission performance. Testing results show that under PCIe Gen3x8, the DMA bandwidth between the hosts and the domestic FPGA accelerator reaches 6.91 GBps (86% utilization rate), supporting up to 16 channels with channel balancing implementation. This design effectively enables large-scale deployment of domestic FPGA heterogeneous accelerators in data center scenarios.

  • Artificial Intelligence
  • Chenyang Li , Chaohong Ma , Xiaofeng Meng
    doi: 10.7544/issn1000-1239.202550195

    The booming of artificial intelligence (AI) plays a crucial role in various fields (e.g., healthcare, bioinformatics, and financial services). The main paradigm of AI applications involves building machine learning models to explore insights and patterns within data, which are then utilized for reasoning and decision-making. The effectiveness and efficiency of AI systems hinge on two key factors: the first is the model aspect (model-centric), involving the enhancement of network structure like the transitioning from RNN to LSTM, and the optimization of model hyperparameters; the second is the data aspect (data-centric), such as standardizing data formats, increasing data size, reducing data noises. For a long time, debugging AI systems primarily focuses on optimizing models. However, the rise of the digital era, characterized by activities such as social networking and e-commerce, has produced vast and varied data sets, making model-centric debugging fall short of fulfilling the requirements of modern applications. Consequently, both research and industry communities have redirected their focus from models to data to address this gap. To this end, “data debugging” has emerged. Rather than tuning model, data debugging places emphasis on scrutinizing data, which understands the impact of data errors on downstream tasks at each stage of machine learning pipeline, and then debug the corresponding errors to improve model performance. On the basis of a comprehensive survey of data debugging related work, the research framework for data debugging is first proposed. According to the interaction between data debugging methods and machine learning pipeline, the existing methods are categorized into closed data debugging, immersive data debugging, and hybrid data debugging. Then, we thoroughly provide a detailed overview of related work in this field. Next, we empirically evaluate data debugging methods, and summarize the commonly used datasets and evaluation metrics in this research area. Finally, the challenges and future directions of data debugging are pointed out.

  • Artificial Intelligence
  • Meixia Wang , Huiying Wang , Miaomiao Huang , Yuhai Zhao
    doi: 10.7544/issn1000-1239.202440643

    Multi-instance multi-label learning (MIML) is an important machine learning framework designed to address complex tasks where each object is represented as a bag of multiple instances and can be associated with multiple labels simultaneously. It has been widely applied in domains such as drug activity prediction and medical image analysis. Different from traditional multi-instance learning and multi-label learning, MIML simultaneously captures the diversity in both the input space (multiple instances) and the output space (multiple labels), enabling more comprehensive modeling and understanding of real-world complex objects. However, this expressiveness also brings significant challenges, including high computational complexity, difficulties in model optimization, and limited generalization capabilities. Therefore, MIML has attracted increasing attention from the research community. Despite this growing interest, a comprehensive survey of MIML is still lacking. We first introduce the formal problem definition and notations related to MIML. Then, from the perspective of data complexity, we categorize MIML problems into three main types: standard MIML, multi-source MIML, and MIML with imprecise labeling. These are further refined into nine subcategories based on solution strategies, instance sources, and label completeness, which are systematically discussed and analyzed. We also summarize commonly used datasets and benchmark comparisons for MIML methods. Finally, we present five representative application scenarios of MIML in real-world tasks, outline four promising future research directions, and conclude the paper.

  • Artificial Intelligence
  • Haiwei Pan , Fengming Yu , Kejia Zhang , Haiyan Lan , Qingyu Meng , Zhe Li
    doi: 10.7544/issn1000-1239.202440694

    Knowledge distillation, as a key technique in deep learning, achieves model compression and acceleration by transferring knowledge from a large teacher model to a smaller student model. Under the premise of maintaining performance, this technology significantly reduces the requirements of computational resources and storage, and facilitates the deployment of high-performance models on resource-constrained edge devices. Firstly, we provide a systematic review of the recent research in knowledge distillation and categorize it from two perspectives: the type of knowledge and teacher-student model architectures. We comprehensively summarize the distillation methods based on three typical types of knowledge: output feature knowledge, intermediate feature knowledge, and relational feature knowledge, as well as distillation methods based on CNN to CNN architecture, CNN to ViT (vision Transformer) architecture, ViT to CNN architecture, and ViT to ViT architecture. Next, we explore various learning paradigms such as offline distillation, online distillation, self-distillation, data-free distillation, multi-teacher distillation, and assistant distillation. Then, we summarize distillation optimization methods based on the distillation process, knowledge structure, temperature coefficient, and loss functions and analyze improvements in distillation brought by adversarial techniques, automated machine learning, reinforcement learning, and diffusion models, and conclude with the implementation of distillation technology in common applications. Despite significant advancements in knowledge distillation, numerous challenges remain in both practical applications and theoretical research. Finally, we provide an in-depth analysis of these issues and offer insights into future development directions.

  • Artificial Intelligence
  • Zituo Li , Jianbin Sun , Guangzhou Chen , Xinyue Fang , Ruijing Cui , Zhiliang Tian , Zhen Huang , Kewei Yang
    doi: 10.7544/issn1000-1239.202550069

    In recent years, large language models (LLMs) have made significant strides in the field of natural language processing (NLP), demonstrating impressive capabilities in both language understanding and generation. However, despite these advancements, LLMs still face numerous challenges in practical applications. One such issue that has garnered extensive attention from both the academic and industrial communities is the problem of hallucinations. Effectively detecting hallucinations in large language models is a critical challenge for ensuring their reliable, secure, and trustworthy application in downstream tasks such as text generation. We provide a comprehensive review of methods for detecting hallucinations in large language models. Firstly, we introduce the concept of large language models and clarify the definition and classification of hallucinations. Then we systematically examine the characteristics of LLMs throughout their entire lifecycle, from construction to deployment, and delve into the mechanisms and causes of hallucinations. Secondly, based on practical application requirements and considering factors such as model transparency in different task scenarios, we categorize hallucination detection methods into two types: those for white-box models and those for black-box models. A focused review and in-depth comparison of these methods are provided. Then we analyze and summarize the current mainstream benchmarks for hallucination detection, laying a foundation for future research in this area. Finally, we identify various potential research directions and new challenges in the detection of hallucinations in large language models.

  • Artificial Intelligence
  • Yiming Li , Yaxin Yu , Zhisheng Yu , Yiting Si , Yusong Ye
    doi: 10.7544/issn1000-1239.202550040

    Multimodal recommendation systems aim to deliver more accurate and personalized recommendations by leveraging diverse data modalities such as text and images. Despite their potential, existing approaches face several critical challenges. 1) Feature distortion. The input embeddings are processed by relatively small pre-trained language models and deep convolutional neural networks, which may lead to inaccurate and incomplete feature representations. 2) Single encoding perspective. The multimodal encoding layers of current models only consider encoding from a single perspective that is either memory-based or expansion-based, which limits representational capacity and leads to information loss. 3) Poor multimodal alignment. Embeddings from different modalities are distributed in distinct and heterogeneous semantic spaces, and need to be effectively mapped into a shared semantic space to achieve accurate alignment. However, existing methods typically rely on simple operations such as direct product of behavioral information, which cannot adequately capture the complex and high-order relationships among modalities. As a result, precise alignment across multiple modalities is difficult to achieve. To address these issues, we propose a novel model named DPRec. This model considers encoding from both memory and expansion perspectives and introduces hypergraphs for multi-level precise cross-modal alignment. The model is evaluated through extensive experiments on three real-world datasets. Experimental results confirm the effectiveness of DPRec, demonstrating its superiority over state-of-the-art approaches in improving recommendation accuracy.

  • Artificial Intelligence
  • Delü Lin , Kun He
    doi: 10.7544/issn1000-1239.202440804

    With the rapid progress of deep learning technology and the continuous exploration of massive datasets, the self-attention module has been widely applied in various fields such as natural language processing, computer vision, and large language models. Although the self-attention module significantly improves the detection accuracy of deep learning models, its huge computational demand makes it particularly difficult to deploy on computing devices with limited computing power. Integer quantization, as one of the key technologies for deploying models on low-power computing chips, faces the problem of high precision loss caused by the structural characteristics of the self-attention module. To address this issue, a thorough analysis of the integer quantization error in the self-attention module is conducted, and two methods, pseudo-softmax vector quantization and block-wise pseudo-softmax vector quantization, are proposed. These two methods aim to significantly improve inference speed while effectively reducing the error caused by integer quantization by performing special integer quantization on the softmax vectors in the self-attention module. Experimental results show that compared with traditional direct quantization methods, the pseudo-softmax vector quantization method can reduce the quantization accuracy loss by 50%, while the block-wise pseudo-softmax vector quantization method can further reduce the accuracy loss by approximately 90%. These results fully demonstrate the effectiveness of the two quantization methods in reducing precision loss, providing strong support for the efficient deployment of the self-attention module on devices with limited computing power.

  • Artificial Intelligence
  • Jiawen Qin , Pengfeng Huang , Qingyun Sun , Haoyi Zhou , Jianxin Li
    doi: 10.7544/issn1000-1239.202440660

    Self-supervised learning has emerged as a promising approach in addressing the issues of label dependency and poor generalization performance in traditional graph neural networks (GNNs). This method leverages the inherent structure and properties of the data to generate supervisory information without relying on extensive labeled datasets. However, most existing self-supervised learning methods hold the assumption that the graphs are homophilic and consequently fail to generalize well to graphs with high heterophily, where connected nodes may have different class labels and dissimilar features. In this paper, we study the problem by developing an asymmetric self-supervised learning on non-homophilic graphs framework (MIRROR), which is easy to implement and does not rely on random graph augmentations and homophilic assumptions. Inspired by the designs of existing heterophilic graph neural networks, MIRROR learns the node representations by capturing one-hop local neighborhood information and informative distant neighbors. Such two properties are obtained through carefully designed pretext tasks that are optimized based on predicted neighborhood context and estimated high-order mutual information. Extensive experiments on various types of graph benchmarks demonstrate that our proposed framework can achieve better performance compared with competitive baselines on non-homophilic graphs. The superiority in multiple downstream tasks also demonstrates the generalization capability of MIRROR.

  • Artificial Intelligence
  • Chaofan Li , Songcan Chen
    doi: 10.7544/issn1000-1239.202550351

    Multi-channel electroencephalography (EEG), as a non-invasive technique, records brain electrical activity through multiple electrodes placed on the scalp, aiding in understanding individual psychological states and assisting in the diagnosis of various diseases. Given the high cost and technical challenges of annotating large-scale EEG data, self-supervised learning (SSL), as a label-free learning paradigm, has garnered widespread attention in the EEG domain. SSL leverages the intrinsic structure of data to learn representations, thereby effectively enhancing the model’s generalization ability. Despite significant advancements in current research, the field still faces the following challenges: First, multi-channel EEG data contain complex spatiotemporal correlations, yet many existing methods are confined to modeling either temporal or spatial dimensions in isolation, rather than integrating both comprehensively, thus limiting the ability to profoundly understand the intrinsic complexities of EEG signals. Second, many current approaches have not effectively integrated segment-level and instance-level information. The former contributes to improving the model’s generalization ability, while the latter aids the model in better adapting to downstream (classification) tasks. To address these challenges, we propose and implement a self-supervised pre-training framework that combines contrastive and reconstruction strategies. Specifically, by introducing a channel masking strategy based on temporal mask reconstruction, our framework effectively captures the spatiotemporal relationships of EEG data. By capturing inter-segment relationships in a fine-grained manner, the framework enhances both the model’s performance and its generalization ability. Meanwhile, integrating instance-level contrastive learning with masked reconstruction tasks helps the model learn representations with instance discriminability and local perceptual ability, thus facilitating better adaptation to downstream tasks. Additionally, the introduction of the self-paced learning mechanism further enhances the model’s generalization ability. Finally, experimental results across multiple EEG tasks demonstrate the effectiveness of the proposed approach.

  • Artificial Intelligence
  • Shuangcheng Wang , Songhao Zhu
    doi: 10.7544/issn1000-1239.202550009

    Few-shot learning is the mainstream method for few-shot object detection, but it has serious shortcomings. 1) The extreme lack of new class samples leads to biased distribution of new class features; 2) Due to the robustness assumption during the fine-tuning process are not necessarily applicable to new class samples, the feature extraction network is unable to extract unbiased new class sample features. To deal with the above two issues, three-stage fine-tuning few-shot object detection method based on cross-module knowledge distillation is proposed. Firstly, the feature distribution calibration strategy is designed to calibrate the feature distribution of new class during the two-step fine-tuning process. Secondly, the proposed first bias reduction strategy, effectively alleviates the bias estimation problem of weight parameters in the linear probing (the first stage of the fine-tuning process), and the proposed inverse first bias reduction effectively alleviates the over-fitting problem of feature extraction network during overall fine-tuning (the second stage of the fine-tuning process). Finally, the proposed cross-module knowledge distillation strategy is utilized to guide the shallow modules of the model to learn deep features to capture more discriminative new class features. A large number of experimental results show that the proposed three-stage fine-tuning few-shot object detection method effectively improves the accuracy and robustness of few-shot object detection.

  • Information Security
  • Futai Zou , Kaida Jiang , Tianyue Zhan , Wei Shi , Liang Zhang
    doi: 10.7544/issn1000-1239.202440871

    Accurately and efficiently identifying security vulnerabilities in Web applications holds significant research value, especially as Web systems grow in complexity and scale. Most existing studies in the field of Web vulnerability discovery have focused on PHP-based applications, rendering them less effective or even inapplicable when transferred to the domain of Java Web applications. Furthermore, traditional vulnerability discovery methods often struggle to meet the demands of large-scale and high-efficiency scenarios. Specifically, these methods face difficulties in achieving the precision of dynamic analysis while maintaining the performance benefits of static analysis. To address these challenges, we propose a novel Web vulnerability discovery approach based on joint frontend-backend analysis. By parsing the frontend code to extract taint source information, the method guides backend analysis for pruning irrelevant paths, thereby enhancing both the vulnerability coverage and discovery efficiency. Additionally, the approach integrates both static and dynamic features of the program to construct a comprehensive code model. It combines techniques such as data flow analysis, taint analysis, symbolic execution, and lightweight dynamic solving to detect and verify potential vulnerabilities. This integrated strategy leads to a significant performance boost with minimal computational overhead. The proposed method is evaluated on 105 Java Web vulnerabilities sourced from CVE (common vulnerabilities and exposure) entries, open-source CMS (content management system) platforms, and community-developed applications. The experimental results demonstrate that each component of the system performs effectively, and the overall framework exhibits strong capability in discovering real-world Java Web vulnerabilities.

  • Information Security
  • Jisheng Dong , Cong Li , Zinan Shen , Daqi Hu , Qingni Shen , Zhonghai Wu
    doi: 10.7544/issn1000-1239.202440830

    Blockchain has received a lot of attention in recent years, and the digital signature scheme is a legitimate way to authenticate transactions in the blockchain. The threshold signature scheme enables the key controlled by a single participant to be managed by multiple participants at the same time, and the transaction can be signed only if more than a threshold number of participants sign at the same time, which expands the application mode and application scenario of cryptocurrency, and reduces the risk of single point of failure. However, existing threshold signature schemes are mostly limited to traditional public key signature algorithms such as ECDSA (elliptic curve digital signature algorithm), which incur significant overhead in public key certificate storage and management. IBC (identity-based cryptosystem) can effectively address this issue, but there is no corresponding research on threshold SM9 signature schemes. Implementing a threshold signature scheme usually requires the use of a two-party secret sharing conversion protocol. To address the low efficiency of the existing two-party secret sharing conversion protocol based on the Paillier encryption algorithm, we propose a two-party secret sharing conversion protocol based on the ElGamal encryption algorithm, and apply it to construct a threshold signature scheme based on SM9. In addition, in order to prove the security of the scheme, an identity-based threshold signature ideal function is constructed, and the security of the scheme is proved under the UC (universally composable) model. Theoretical analysis and experimental results show that the threshold SM9 signature scheme based on the ElGamal algorithm is more efficient than the existing schemes. In particular, when the threshold is 100, the speed is about 6.1 times that of the Paillier-based scheme. The threshold signature scheme is tested on three different blockchain platforms, Ethereum, Hyperledger Fabric and ChainMaker on Tencent blockchain as a service. The experimental results show that the scheme is feasible in the blockchain scenario.

  • Information Security
  • Zhipeng Gao , Yi Zhang , Weijing You , Ze Chai , Yang Yang , Lanlan Rui
    doi: 10.7544/issn1000-1239.202440687

    Diffusion models have gained significant attention in recent years due to their potential in various generation tasks, including image and text generation. However, the widespread use of these models has also raised concerns regarding data privacy, particularly the vulnerability of these models to membership inference attacks (MIA). These attacks aim to determine whether a specific data point is part of the model’s training set, posing significant risks to privacy. We provide an overview of the latest developments in privacy protection for diffusion models, with a specific focus on MIAs and their challenges. Existing MIA methods often struggle with balancing the effectiveness of short-term and long-term attacks, and their applicability to diffusion models has not been thoroughly explored. To address these issues, we propose a novel temporal membership inference attack method designed to enhance the attack success rate (ASR) for both short-term and long-term attacks. The proposed method leverages gradient information from noise during short-term attacks and temporal noise patterns to bolster the effectiveness of long-term attacks. Experimental results demonstrate that our method improves the ASR by approximately 5% for short-term attacks and 1% for long-term attacks compared with conventional approaches on common diffusion models. This work contributes to the ongoing efforts to understand and mitigate privacy risks in diffusion model applications.

  • Information Security
  • Peiyun Zhang , Shuai Tao , Jian Chen , Zihan Chen
    doi: 10.7544/issn1000-1239.202440972

    With the rapid development of blockchain technology, assets such as data are no longer satisfied with circulating only on a single blockchain, so cross-chain technology has emerged. Cross-chain technology has improved the problem of blockchain data islands, but it has also brought new challenges to the transmission and sharing of data on different blockchains. On the one hand, different blockchain users usually store data in different ways, and the data verification methods of each blockchain are often not unified. Therefore, users on other blockchains may transmit false or inconsistent data, making it difficult to effectively verify the authenticity and consistency of the data, thereby affecting the reliability of cross-chain access control. On the other hand, although the current proxy re-encryption technology can effectively implement one-to-one access control, a single proxy has a centralization problem and faces the problem of high access control time overhead caused by multi-user concurrency in cross-chain scenarios. In response to the above problems, a two-stage cross-chain access control mechanism in the relay chain environment is proposed, which is designed from two stages: data verification and proxy re-encryption. In the data verification stage, to ensure the authenticity and consistency of the source of encrypted data, combined with the Fiat-Shamir heuristic and IPFS, a data verification method is designed to quickly verify the data, so that the encrypted data can be correctly received by the user, and the encrypted data is securely stored in IPFS to relieve the storage pressure on the chain. In the proxy re-encryption stage, an improved threshold proxy re-encryption method is designed to effectively deal with access control in multi-user scenarios, and the re-encryption process is optimized based on the re-encryption update algorithm. While achieving decentralized re-encryption, the time overhead of re-encryption is reduced, and the efficiency of re-encryption transaction processing for multi-user access control is effectively improved. Experimental results show that the proposed mechanism has a small overhead in data verification, and the overhead of encryption, re-encryption, and decryption processes in multi-user scenarios is better than that of the comparative methods.