Home Latest Articles
Latest Articles
  • Pengfei Jiao, Shuxin Chen, Xuan Guo, Dongxiao He, Dong Liu
    Journal of Computer Research and Development. 2024, 61(8): 2045-2066.

    Graph neural networks (GNNs) are powerful tools for handling graph-structured data, capable of capturing complex relationships and features among nodes. However, the discrete architecture of GNNs leads to numerous challenges in representing graph structures, modeling graph evolution, adapting to irregular data, and managing computational costs. In response to these challenges, neural ordinary differential equations (ODEs) have been introduced as a novel method to address the challenges faced by GNNs, as they can simulate the continuous evolution of system states, providing continuous deep encoding and inference capabilities. However, neural ODEs are designed for Euclidean structured data and cannot directly capture the characteristics of graphs. Therefore, researchers have proposed graph neural ODEs, a new type of architectures that combines neural ODEs with GNNs, which can better adapt to graph-structured data and fully utilize its characteristics. In recent years, research related to graph neural ODEs has delved into various directions of graph machine learning, sparking a new research trend. In this context, we systematically review the relevant research of graph neural ODEs in a timely manner. Firstly, we review the key advantages of GNNs and the challenges they face, and elucidate the theoretical basis and practical significance of introducing neural ODEs and combining them with GNNs. Subsequently, we elaborate on the background and basic concepts of graph neural ODEs, proposing a novel taxonomy, and comprehensively describe some important current methods on the taxonomy. Then, we introduce commonly used verification methods in related research, including downstream tasks and datasets. Furthermore, we delve into the applications of graph neural ODEs in multiple practical fields. Finally, we summarize and prospect the challenges and future development trends of graph neural ODEs.

  • Zhuang Liu, Zichen Dong, Yilin Dong, Jiaming Shang, Fan Zhang, Yuran Chen, Peiyan Lou, Xinran Sun, Yu Wang, Jun Zhao, Lin Wayne
    Journal of Computer Research and Development. 2024, 61(8): 2067-2096.

    Lifelong Graph Learning (LGL) is an emerging field aimed at achieving continuous learning on graph-structured data, addressing the catastrophic forgetting issue in existing tasks and enabling models to adapt to newly emerging graph tasks in a sequential manner. Despite demonstrating strong learning capabilities, improving LGL's sustained performance remains a crucial challenge. To address the gaps in existing research, we provide a comprehensive survey and summary of recent developments in the field of LGL. Firstly, we reclassify existing methods of LGL, focusing particularly on approaches to overcome catastrophic forgetting. Secondly, we systematically analyze the strengths and weaknesses of these methods, and discuss potential solutions for achieving sustained performance improvements. Our study emphasizes on how to avoid forgetting old tasks during the continual learning process while swiftly adapting to the challenges of new tasks. Finally, we also discuss the future directions of LGL, covering potential impacts in the application domains, open issues, and specifically analyze their potential effects on sustained performance improvements. These discussions will help guide future research directions in LGL, promoting further development and application in this field.

  • Chuang Wang, Yan Ding, Chenlin Huang, Liantao Song
    Journal of Computer Research and Development. 2024, 61(8): 2097-2109.

    SM4 algorithm is a commercial block cipher algorithm independently designed by China, and its encryption and decryption performance has become one of the critical factors affecting the data confidentiality of the information system. The existing optimizations mainly focus on hardware designs and software look-up tables, which have problems such as dependence on specific hardware environments, low efficiency, and vulnerability to side-channel attacks. Bit slicing technology efficiently processes block ciphers in parallel by reorganizing input data, and can resist side-channel attacks against caches. However, the existing researches on bitsliced block ciphers are highly dependent on the hardware platforms and only support a single processor architecture, and the parallel processing pipeline starts slowly. It is difficult for the encryption and decryption operations for small-scale data to give full play to the advantages of advanced instruction sets such as SIMD (single instruction multiple data) instructions. To resolve the above problems, we firstly propose a cross-platform general bitsliced block cipher algorithm model, which supports a general data slicing method that provides consistent data slicing for different processor instructions. Based on that, a fine-grained bitsliced SM4 optimization algorithm for SIMD instructions is proposed, which can effectively shorten the startup time of the algorithm through fine-grained plaintext slicing reorganization and linear transformation optimization. The experiments show that, compared with the look-up table-based SM4 algorithm, the encryption rate can reach up to 438.0 MBps. The clock cycles required for encrypting a byte are up to 7.0 CPB (cycle/B), and the encryption performance is improved by an average of 80.4% to 430.3%.

  • Haibo Wu, Hui Liu, Yi Sun, Jun Li
    Journal of Computer Research and Development. 2024, 61(8): 2110-2126.

    With the prevalence of blockchain technology, Hyperledger Fabric (Fabric for short), as a well-known open source blockchain platform, has received wide attention. However, Fabric still suffers from conflicts between concurrent transactions. Conflicts will cause a large number of invalid transactions entering the chain, resulting in a decrease in throughput and hindering its development. For this problem, existing intra-block-conflict-oriented schemes lack efficient conflict detection and avoidance methods, and ignore the adverse impact of inter-block conflicts on throughput. We propose an optimization scheme for Fabric, Fabric-HT (Fabric with high throughput), from both intra-block and inter-block aspects to effectively reduce concurrency inter-transaction conflicts and improve system throughput. For intra-block transaction conflicts, we present a transaction scheduling mechanism, in which an efficient data structure (the dependency chain) is defined to identify and abort transactions with “dangerous structures” in advance, reasonably schedule transactions and eliminate conflicts; for inter-block transaction conflicts, the conflict transaction detection is moved to the sorting node to complete, and an early conflict transaction avoidance mechanism following “push-match” pattern is established. A large number of experiments are carried out in multiple scenarios, and the results show that Fabric-HT overperforms existing schemes in terms of throughput, transaction abort rate, average transaction execution time, and invalid transaction space occupancy. The results show that the throughput of Fabric-HT can reach up to 9.51x that of Fabric and 1.18x of the latest optimized scheme FabricSharp; compared with FabricSharp, the space utilization is increased by 14%. In addition, Fabric-HT also shows good robustness and anti-attack ability in solving concurrent transaction conflict.

  • Yuan Shen, Wei Song, Changsheng Zhao, Zhiyong Peng
    Journal of Computer Research and Development. 2024, 61(7): 1611-1628.

    As a widely used ciphertext authorization access mechanism in cloud environments, ciphertext-policy attribute-based encryption (CP-ABE) has fine-grained, one-to-many and owner-controlled properties. However, the traditional CP-ABE mechanism is difficult to obtain the identities of authorized users who maliciously abuse their decryption privileges since multiple users may have the same attribute set. Although numerous existing studies achieve the identity tracking for some specific decryption privilege abuses (i.e., white-box attacks and black-box attacks), they are challenging to audit authorized users’ identities for ciphertext access behaviors, which may lead to potential data security and owners’ right-to-be-informed compliance issues. Based on CP-ABE mechanism, to realize identity tracing of ciphertext data access behavior in real application scenarios, this scheme designs a cross-domain ciphertext data sharing method, which generates the access request by binding the traceable decryption key with the authorized user’s access behavior. The integrity of access requests is protected by blockchain. Meanwhile, this scheme introduces an encrypted inverted index structure to address the inefficiency of the identity traceability caused by blockchain traversal. The privacy-preserving of index queries is achieved through the BLS signature and privacy set intersection. Theoretical analysis and experimental results demonstrate that the proposed cross-domain ciphertext sharing scheme with authorized users’ access behaviors audit trail is efficient and practical.

  • Jihong Wang, Shuqing Zhao, Minnan Luo, Huan Liu, Xiang Zhao, Qinghua Zheng
    Journal of Computer Research and Development. 2024, 61(7): 1629-1642.

    Misinformation detection is crucial for the social stability. Researches show that there are substantial distinctions between misinformation and real information in terms of information content and propagation structure. Consequently, recent researchers mainly focus on improving the accuracy of misinformation detection by jointly considering the information content and propagation structure. However, these methods can be infeasible in practice since they highly rely on manual label information. The manual labels can be expensive since they require extensive comparison with official reports and other evidence. Moreover, the spreaders of misinformation can adversarially manipulate the information content and propagation structure by controlling reviews and other methods. Such behaviors may exacerbate the challenges of misinformation detection. To address these problems, we propose a robust few-label misinformation detection method based on information bottleneck theory. Specifically, to mitigate the dependence on labeled data, we propose to integrate the unlabeled sample information by employing the mutual information maximization technique. Furthermore, to improve the robustness of our method against the adversarial manipulation of misinformation spreaders, we employ the adversarial training strategy to simulate the behaviors of the spreaders and propose to learn robust representations based on the information bottleneck theory. The learned representations can effectively embed the essential information in the misinformation while discarding the adversarial information involved by the spreaders. Empirical evaluations validate the effectiveness of the proposed approach, demonstrating superior performance compared with benchmark methods in terms of few-label detection and robustness.

  • Peiwang Fu, Hongfa Ding, Hai Liu, Heling Jiang, Mingli Tang, Yingying Yu
    Journal of Computer Research and Development. 2024, 61(7): 1643-1669.

    Mass distributed graph data, generated by social network, social IoT (Internet of things) and other scenarios, are collected by the service providers and used to provide diverse data oriented services, and they arise serious privacy conserns. In this context, how to achieve secure and effective collecting of distributed graph-structure data with strong correlation has become a bottleneck in large-scale graph-structure data application services. The Node/Edge-LDP (local differential privacy) model of distributed graph-structure data cannot effectively handle the conflict between effectiveness of privacy preserving and the utility of data. To this end, a statistics collecting algorithm named GS-LDP of distributed graph via LDP is proposed. This algorithm can collect three different statistics, i.e., degree distribution, triangle count sequence and clustering coefficient, simultaneously. And it can adapt to different effectiveness and privacy preserving requirements. First, a degree distribution collecting algorithm with Node-LDP is designed. By using the grouping mechanism and symmetric unary coding mechanism, this algorithm can achieve high strength privacy protection. Second, obtaining the threshold value with the algorithm mentioned above, then alleviating the problem of excessive noisy edges in random noise through introducing a prune algorithm, the triangle count sequence collecting algorithms with Node-LDP and Edge-LDP are proposed respectively. Third, based on the aforementioned algorithms, clustering coefficient collecting algorithms with Node-LDP and Edge-LDP are proposed respectively, by introducing the Laplace mechanism. And these algorithms achieve multi-metrics collecting of distributed graph-structure for different privacy and utility requirements. At last, experimental and comparative results show that, the proposed algorithms can strengthen both privacy and data utility, and perform significantly better than existing single or multiple statistics collecting algorithms.

  • Yudong Cheng, Fang Zhou
    Journal of Computer Research and Development. 2024, 61(7): 1670-1680.

    Anomaly detection aims to identify data that deviates from expected behavior patterns. Despite the potential of semi-supervised anomaly detection methods in enhancing detection accuracy by utilizing a limited amount of labeled data as prior knowledge, the labeled anomalies (i.e., seen anomalies) acquired are unlikely to cover all types of anomalies. In real-world scenarios, novel types of anomalies (i.e., unseen anomalies) often emerge, which may exhibit distinct characteristics from the known anomalies, thereby rendering them challenging to detect using existing semi-supervised anomaly detection methods. To address this issue, we propose a semi-supervised unknown anomaly detection (SSUAD) method, aimed at simultaneously identifying both known and unseen anomalies. This method utilizes a closed-set classifier for the classification of known anomalies and normal instances, and an unknown anomaly detector for the detection of unseen anomalies. Moreover, considering the extreme imbalance between anomalies and normal instances in the anomaly detection scenario, we design an effective data augmentation strategy to increase the number of anomaly samples. Experiments are conducted on UNSW-NB15 and KDDCUP99 datasets, as well as a real-world dataset SQB. The results reveal that, compared with existing anomaly detection methods, SSUAD exhibits significant improvement in the anomaly detection performance metrics AUC-ROC and AUC-PR, thereby verifying the effectiveness and reasonableness of the proposed method.

  • Yunfan Gao, Dongqing Yu, Siqi Wang, Haofen Wang
    Journal of Computer Research and Development. 2024, 61(7): 1681-1696.

    Site selection, as a core link in business decisions and urban infrastructure planning, plays a significant role in whether physical stores and urban infrastructure can perform as expected. The existing site selection recommender system’s data service orchestration is rather fixed and unable to make timely adjustments to different user needs. Its application scenarios are limited, and the system’s flexibility and scalability in human-computer interaction are poor. Recently, large language models (LLMs) such as GPT-4 have demonstrated powerful capabilities in intent understanding, task orchestration, code generation, and tool usage. They can accomplish tasks that traditional recommendation models struggle to balance, providing new opportunities for reshaping the recommendation process and implementing integrated recommendation services. However, site selection recommendation not only shares the common challenges of traditional recommendations but also presents unique challenges due to its reliance on spatial data. Against this backdrop, we propose a LLM-powered site selection recommendation system. Firstly, we expand the scenarios of site selection recommendation and propose a scene recommendation task of finding suitable store types based on location, combining collaborative filtering algorithms and spatial pre-training models. Secondly, we construct a site selection decision engine driven by a large language model. The language model itself has many defects in dealing with space-related tasks, such as the lack of spatial awareness, inability to understand specific locations, and the tendency to fabricate place names and addresses. In this paper, we propose a mechanism for handling spatial tasks within the language model framework. By utilizing geocoding, reverse geocoding, and address resolution tools, we enhance the model’s spatial awareness and prevent address fabrication issues. In combination with site selection recommendation models, scenario recommendation models, external knowledge bases, and map visualization, we accomplish diverse tasks in site selection recommendations. We achieve intelligent planning, execution, and attribution for site selection tasks, enhance the interaction experience of spatial service systems, and provide new design and implementation ideas for future AI-driven site selection recommender systems.

  • Beilun Wang, Jiaqi Zhang, Yinghao Cai, Zhaoyang Wang, Xiao Tan, Dian Shen
    Journal of Computer Research and Development. 2024, 61(7): 1697-1712.

    Tensor data (or multi-dimensional array data) are often generated in information systems of various industries, such as functional magnetic resonance imaging (fMRI) data in medicine systems and user-product data in product information systems. By using these data to predict the relationship between tensor features and univariate responses, data empowerment can be achieved, providing more accurate services or solutions, such as disease decision diagnosis or product recommendations. Currently available tensor regression methods, however, present two major shortcomings: the spatial information of tensors may be lost in these models, resulting in inaccurate prediction results; the calculation cost is too high, which results in untimely solutions or services. The two problems are more severe for large-scale data with high-order structures. Therefore, in order to achieve data empowerment, that is, to use tensor data to improve the quality and efficiency of information services or solutions, we propose sparse and low-rank tensor regression model (SLTR). This model enforces sparsity and low-rankness of the tensor coefficient by directly applying $ {\mathscr{l}}_{1} $ norm and tensor nuclear norm on it respectively, such that not only the structural information of the tensor is preserved but also the data interpretation is convenient. To make the solving procedure scalable and efficient, SLTR makes use of the proximal gradient method to optimize the hybrid regularizer, which can be easily implemented parallelly. Additionally, a tight error bound of SLTR is theoretically proved. We evaluate SLTR on several simulated datasets and one video dataset. Experimental results show that, compared with previous models, SLTR is capable to obtain a better solution with much fewer time costs.