Home Latest Articles
Latest Articles
  • Zetao Jiang, Wencai Zhu, Xin Jin, Peiqi Liao, Jingfan Huang
    Journal of Computer Research and Development. 2024, 61(11): 2897-2908.

    As visual features closer to the text domain, the grid features extracted by the CLIP (contrastive language-image pre-training) image encoder are easy to convert into the corresponding semantic natural language, which can alleviate the semantic gap problem, so it may become an important source of visual features in the image captioning in the future. However, this method does not consider that the division of image content may cause a complete object to be divided into several grids. The segmentation of the objects will inevitably lead to the lack of a complete expression of the object information in the feature extraction results, and further lead to the lack of an accurate expression of the object and the relationship between the objects in the generated sentence. Aiming at the phenomenon of grid features extracted by CLIP image encoder, we propose dual semantic collaborative network (DSC-Net) for image captioning. Specifically, dual semantic collaborative self-attention (DSCS) module is first proposed to enhance the expression of object information by CLIP grid features. Then dual semantic collaborative cross-attention (DSCC) module is proposed to integrate semantic information between grid and object to generate visual features, and to be used to predict sentences. Finally, dual semantic fusion (DSF) module is proposed to provide region-oriented fusion features for the above two semantic cooperation modules, and to solve the problem of correlation conflicts that may arise in the process of semantic cooperation. After a large number of experiments on the COCO dataset, the proposed model achieves a CIDEr score of 138.5% on the offline test set divided by Karpathy et al., and a CIDEr score of 137.6% in the official online test. Compared with the current mainstream image captioning methods, this result has obvious advantages.

  • Chen Li, Yidong Chen, Zhonghua Lu, Xueying Yang, Zitian Wang, Xuebin Chi
    Journal of Computer Research and Development. 2024, 61(11): 2909-2922.

    Multi-objective optimization problems are common in real-world applications. However, the high computational complexity severely hinders their solution. Currently, multi-objective evolutionary algorithms are mostly used for their solving. Nevertheless, these methods usually contain random operations in the process of population initialization and evolution to maintain diversity, which leads to non-reproducibility and lacking the theoretical guarantee of global convergence. Therefore, a novel multi-objective Dividing Rectangles(DIRECT) algorithm is introduced based on decomposition strategy. In this algorithm, the original problem is decomposed into a series of subproblems by a normalized decomposition method that can better capture the complex frontier to reduce the computational complexity of the problem. Meanwhile, the Dividing Rectangles algorithm is used to simultaneously optimize the decomposed subproblems, and the generated candidate solutions are assigned to the corresponding subproblems based on the global association mechanism in the optimization process to better retain the good candidate solutions and improve the algorithm search efficiency. Finally, the convergence of the algorithm is demonstrated. To further improve the computational efficiency, a multi-level and multi-granularity parallel scheme based on the adaptive associative migration strategy is proposed, based on which the proposed algorithm is parallelized. Furthermore, the proposed algorithm and its parallel version are applied to several benchmark optimization problems, and the experimental results show that the proposed algorithm can obtain a Pareto set with better convergence and diversity than NSGA-II, and the parallel version can further improve the accuracy of Pareto frontier approximation while massively reducing the problem-solving time.

  • Xiao Kang, Xingbo Liu, Pengyu Lu, Zhijie Zhao, Xiushan Nie, Shaohua Wang, Yilong Yin
    Journal of Computer Research and Development. 2024, 61(11): 2923-2936.

    Online cross-modal hashing has received increasing attention due to its efficiency and effectiveness in handling cross-modal streaming data retrieval. Despite promising progress, most existing methods rely on accurate and clear supervised information. There are only a few online cross-modal hashing studies concentrating on unsupervised learning mode, and numerous challenges still need to be tackled. For example, streaming data usually suffer from the unbalanced distribution problem due to the limited volume of data in each chunk. Most existing methods neglect this problem, resulting in heightened sensitivity to outlier samples and compromised robustness. Moreover, existing models typically exploit global data distribution, while ignoring local neighborhood information that can promote hash learning. To solve these problems, we propose an unsupervised online cross-modal hashing method with double structure-preserving, called SPOCH(structure preserving online cross-modal hashing). It simultaneously explores the global structure and local structure to generate the corresponding common representation; thereafter, the learned common representation can be used to guide the hash learning process. In terms of global structure-preserving, we design the loss function based on $ {L}_{2,1} $ norm, which can alleviate the sensitivity of outlier samples. In terms of local structure-preserving, we reconstruct sample representation based on neighbor relations that integrates the multi-modality information. In addition, to alleviate the forgetting problem, we propose joint optimization on streaming data, and design the corresponding update strategy to improve the training efficiency. We conduct experiments on two widely-used cross-modal retrieval datasets. Compared with the existing state-of-the-art unsupervised online cross-modal hashing methods, SPOCH achieves superior retrieval accuracy within a comparable or even shorter training time, validating the effectiveness of the proposed approach.

  • Zepeng Wang, Chao Ma, Zhuangzhuang Zhang, Libing Wu, Xiaochuan Shi
    Journal of Computer Research and Development. 2024, 61(10): 2404-2416.

    In recent years, the industrial control network has been developing rapidly. The advantages of digitization, intelligence, and automation have brought significant benefits to the industry while also introducing increasingly complex and variable attack threats. In the context of data element security, timely detection and response to industrial control network threats have become an urgent task to be solved. By continuously monitoring and analyzing the data flow in industrial control networks, the problem of industrial control network threat detection can be transformed into a time series anomaly detection problem. However, the existing time-series anomaly detection methods are limited by the quality of industrial control network datasets and are often sensitive to only a single type of anomaly while ignoring other anomalies. Therefore, in this paper, we propose a deep reinforcement learning and data augmentation based threat detection method in industrial control networks (DELTA). DELTA introduces a novel data augmentation selection technique for time series datasets, which allows for the selection of appropriate data augmentation operations sets tailored to different baseline models to enhance the quality of the industrial control network time series datasets. Simultaneously, deep reinforcement learning algorithms (A2C/PPO) dynamically select candidate models from the baseline models at different time points, leveraging multiple types of anomaly detection models to address the issue of sensitivity to single-type anomalies. The experimental results compared with the existing time series anomaly detection models show that DELTA has a significant improvement in accuracy and F1 value over all baseline models at an acceptable cost of additional time consumption, which verifies the effectiveness and practicality of the method.

  • Haobo Li, Mohan Li, Peng Chen, Yanbin Sun, Zhihong Tian
    Journal of Computer Research and Development. 2024, 61(10): 2417-2432.

    Data identification is a prerequisite for achieving precise data governance, effectively ensuring the security of data elements during cross-domain transfer. Currently, there are methods for generating identifiers for individual data, but as the scale of data continues to expand, identifiers at the data level cannot be directly applied to the dataset level. This also introduces issues of identifiers being “easily damaged” and “difficult to embed”. To effectively address these issues, we adopt the design concept of network honeypoint from the “guardian” model proposed by academician Fang Binxing. Utilizing the idea of deception defense, we propose an anti-damage data identification technology based on dataset honeypoint for cross-domain data transfer scenarios, and design a complete method for generating and embedding dataset honeypoints. First, for cross-domain data transfer scenarios, dataset honeypoints are designed. By enhancing the concealment of dataset honeypoints and increasing their redundancy, the issue of identifiers being “easily damaged” is addressed. Second, by ensuring that the form of dataset honeypoint is indistinguishable from real data, the issue of identifiers being “difficult to embed” is resolved. Finally, experiments conducted on both image and encrypted text data modalities demonstrate that dataset honeypoints possess high anti-damage capability, high robustness, and low performance overhead.

  • Xiujun Wang, Lei Mo, Xiao Zheng, Linna Wei, Jun Dong, Zhi Liu, Longkun Guo
    Journal of Computer Research and Development. 2024, 61(10): 2433-2447.

    Many cloud native database applications need to handle massive data streams. To analyze group trend information in these data streams in real time without compromising individual user privacy, these applications require the capability to quickly create differentially private histograms for the most recent dataset at any given moment. However, existing histogram publishing methods lack efficient data structures, making it difficult to rapidly extract key information to ensure real-time data usability. To address this issue, we deeply analyze the relationship between data sampling and privacy protection, and propose a sampling based fast publishing algorithm with differential privacy for data stream (SPF). SPF introduces an efficient data stream sampling sketch structure (EDS) for the first time, which samples and statistically estimates data within a sliding window and filters out unreasonable data, enabling rapid extraction of key information. Then, we demonstrate that the approximations output by the EDS structure are theoretically equivalent to adding differential privacy noise to the true values. Finally, to meet the privacy protection strength provided by the user while reflecting the true situation of the original data stream, an adaptive noise addition algorithm based on efficient data stream sampling is proposed. According to the relationship between the user-provided privacy protection strength and the privacy protection strength provided by the EDS structure, the algorithm adaptively generates the final publishable histogram through privacy allocation. Experiments show that compared with existing algorithms, SPF significantly reduces time and space overhead while maintaining the same data usability.

  • Yunshu Hu, Jun Zhou, Zhenfu Cao, Xiaolei Dong
    Journal of Computer Research and Development. 2024, 61(10): 2448-2466.

    As the development of the emerging areas of network services such as big data and cloud computing, data element has played an increasingly critical role in the fields of intelligent e-health and scientific research. Gene sequencing technology is widely used in many fields to determine the cause and category of a patient’s disease, by processing the patient’s gene sequence. Due to constrained storage and computing resources, local users often need to rent resource-abundant cloud servers, unfortunately always working in untrusted environments, to fulfill the computationally-intensive task of large-scale gene sequencing function evaluation. To guarantee users’ data privacy and the correctness of computing results, most of the state-of-the-art methods exploit the techniques of public key fully homomorphic encryption and secure multiparty computation to achieve data privacy, and the technique of Yao’s garbled circuit or bilinear paring to achieve correctness verification. Owing to the fact that huge computational overhead and communication overhead are required in the cryptographic primitives mentioned above, they are inappropriate for efficiency needs of the resource-constrained local users in gene sequence analysis. To address this challenging issue, in this paper, a lightweight verifiable privacy-preserving gene sequence analysis scheme in the multi-user setting is proposed. Firstly, we design an efficient verifiable multi-key homomorphic data encapsulation mechanism VMK-HDEM. The proposed VMK-HDEM enables batch outsourced function evaluation on L different input instances over the encrypted domain. The usage time complexity of public key encryption on each user $ {S en}_{i} $’s end is O(L), independent of the dataset size $ {n}_{i} $, which significantly decreases the computational cost of local users. For verification, the size of the proof is $ O({deg}_{F}) $ where $ {deg}_{F} $ denotes the degree of the function, independent to the size of user’s dataset $ {n}_{i} $. Furthermore, based on our constructed cryptographic primitive VMK-HDEM, a lightweight and efficient verifiable privacy-preserving gene sequence analysis scheme LWPPGS is proposed. It not only can preserve the privacy of both users’ gene datasets and the results of gene sequence analysis, but also efficiently verify the correctness of the outcome. Finally, formal security proof and experimental simulation results show the security and practicability of our proposed VMK-HDEM and LWPPGS.

  • Bing Han, Hao Wang, Min Fang, Yongchao Zhang, Lu Zhou, Chunpeng Ge
    Journal of Computer Research and Development. 2024, 61(10): 2467-2481.

    Lightweight mobile devices with limited resources often alleviate their computational and storage burdens by outsourcing large-scale data to cloud storage servers. However, this cloud storage model is susceptible to the possibility of selfish cloud servers discarding data to conserve storage resources. Therefore, there is a need for effective integrity verification of cloud-stored data to ensure its correct and intact storage. Existing cloud storage integrity verification mechanisms lack a reliable approach to perform real-time, multiple verifications of data under the premise of data privacy protection. We propose an integrity verification mechanism based on a trusted execution environment. It generates trustworthy proofs in isolated areas to ensure that the cloud server remains unaware of the data and the entire proof generation process, thereby compelling honest assurance of data integrity throughout the process. To further enhance the security of the proposed solution, we introduce blockchain smart contracts to provide trustworthy storage and verification of proofs. Additionally, we address the issue of resource scarcity on the client side by proposing an efficient verification mechanism based on cuckoo filters. Experimental results demonstrate that this method can achieve high execution efficiency and practicality while ensuring the integrity verification of private data.

  • Wei Li, Wei Feng, Yu Qin, Dengguo Feng
    Journal of Computer Research and Development. 2024, 61(10): 2482-2500.

    Confidential computing (CC) is based on the hardware TEE, which protects the confidentiality and integrity of data in use through isolation, integrity measurement, and remote attestation, and is protected from attacks by privileged adversaries. However, the measurement and attestation mechanisms of existing CC platforms focus on the launch time integrity instead of the runtime. When potential memory vulnerabilities in user workloads are exploited by adversaries, it can lead to attacks such as control flow hijacking. The existing CC platforms that only protect integrity at startup cannot effectively prevent or detect such runtime attacks. To address that, we propose a scheme of runtime monitoring in CC based on dynamic integrity measurement. By introducing control and data flow measurements into the TEE and verifying the measurements by a trusted verifier, the remote attestation and runtime integrity protection are achieved on the CC platform. The implementation of our prototype on CSV/SEV servers with confidential VM/container architecture shows that the scheme achieves higher runtime security with about 16% performance overhead.

  • Wenpeng Song, Liang Zhang, Yuhang Ma
    Journal of Computer Research and Development. 2024, 61(10): 2501-2513.

    The continuous rapid development of Internet technology has brought increasing convenience for data sharing. However, data security and privacy issues have also emerged. Taking healthcare as an example, people hope to seek treatment quickly and claim reimbursement promptly, but do not want to disclose their medical records and expenses in hospitals. To adapt to the above data sharing scenarios, we propose a decentralized identity (DID) model based on blockchain and proxy re-encryption (PRE). Further, we achieve a data sharing framework for smart healthcare. Blockchain is leveraged to avoid the problem of single-node failure, ensuring data availability and data consistency. PRE is employed to achieve the separation of data generation, management and usage. This character is identical to that of the DID model, where verifiable credential (VC) issuance and verification are separated. Besides, PRE helps to achieve the goal of patient-centered healthcare. Furthermore, we obtain verifiable presentation (VP) by using the BLS aggregation signatures, which enable us to combine and authenticate multiple VCs. We incorporate blockchain as a decentralized trusted third party to check the correctness of cryptographic operations, achieving data verifiability and audibility in an encrypted mode. Also, we make a scrupulous security analysis of the proposed framework and compare it with related work. Finally, we conduct comprehensive experiments based on Ethereum and IPFS (inter-planetary file system), demonstrating the feasibility and efficiency of our solution.