Search

With the rapid development of network technology, the Internet has brought significant convenience to various sectors of society, holding a prominent position. Due to the unpredictable and severe consequences resulting from malicious attacks, the detection of anomalous network traffic has garnered considerable attention from researchers over the past few decades. Accurately labeling a sufficient amount of network traffic data as a training dataset within a short period of time is a challenging task, given the rapid and massive generation of network traffic data. Furthermore, the proportion of malicious attack traffic is relatively small compared to the overall traffic data, and the distribution of traffic data across different types of malicious attacks also varies significantly. To address the aforementioned challenges, this paper presents a novel network anomaly detection algorithm based on semi-supervised learning and adaptive multiclass balancing. Building upon the assumption of consistent distribution between labeled and unlabeled data, this paper introduces the multiclass split balancing strategy and the adaptive confidence threshold function. These innovative approaches aim to tackle the issue of the multiclass imbalanced in traffic data. By leveraging the mutually beneficial relationship between semi-supervised learning and ensemble learning, this paper presents the collaborative rotation forest algorithm. This algorithm is specifically designed to enhance performance of anomaly detection in an environment with label inadequacy. Several comparative experiments conducted on the NSL-KDD, UNSW-NB15, and ToN-IoT demonstrate that the proposed algorithm achieves significant improvements in performance. Specifically, it enhances precision by 1.5–5.7%, recall by 1.5−5.7%, and F-Measure by 1.4−4.3% compared to the state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

Network intrusion detection system plays a crucial role in protecting the integrity and availability of sensitive assets, where the detected traffic data contain a large amount of time, space, and statistical information. However, existing research lacks the utilization of spatial-temporal multi-granularity data features and the mutual support among different data features, thus making it difficult to specifically and accurately identify anomalies. Considering the distinctions among different granularities, we propose a framework called tri-broad learning system (TBLS), which can learn and integrate the three granular features. To explore the spatial-temporal connotation of the traffic information accurately, a feature dataset containing three granularities is constructed according to the characteristics of time, space, and data content. In this way, we use broad learning basic units to extract abstract features of different granularities and then express these features in different feature spaces to enhance them separately. We use a normal distribution initialization method in BLS to optimize the weights of feature nodes and enhancement nodes for better detection accuracy. The merits of our proposed model are exhibited on the UNSW-NB15, CIC-IDS-2017, CIC-DDoS-2019, and mixed traffic datasets. Experimental results show that TBLS outperforms the typical BLS in terms of various evaluation metrics and time consumption. Compared with other machine learning methods, TBLS achieves better performance metrics. [ABSTRACT FROM AUTHOR]

Network intrusion detection plays an important role as tools for managing and identifying potential threats, which presents various challenges. Redundant features and difficult marking in data cause a long-term problem in network traffic detection. In this paper, we propose a semi-supervised machine learning framework based on multi-strategy feature filtering, principal component analysis (PCA), and an improved Tri-Light Gradient Boosting Machine (Tri-LightGBM) based on stratified sampling. This multi-strategy feature filtering method employing Fisher score and Information gain can select features that have good category discrimination and are more relevant to category labels. After that, we combine PCA to convert multiple features into comprehensive features, which are used as the input of the Tri-LightGBM model. Tri-LightGBM can exploit unlabeled data cooperatively and maintain a large disagreement among the base learners. Moreover, we propose a stratified sampling based on labeled categories to reduce the probability of being selected as the same category during the model update process. Thus, the Tri-LightGBM based on stratified sampling can compensate for the classification error rate caused by the imbalance of the dataset. The semi-supervised machine learning framework is evaluated on two intrusion detection evaluation datasets, namely UNSW-NB15 and CIC-IDS-2017. The evaluation results show that the multi-strategy feature filtering method can increase the accuracy, recall, precision, and F-measure by up to 0.5%, and reduce the false-positive rate by up to 0.5%. Furthermore, the precision rate of minority categories can be increased by about 1–2%. [ABSTRACT FROM AUTHOR]

The Harris hawks optimization (HHO) algorithm is a new metaheuristic algorithm proposed in recent years. Due to the shortcomings of this algorithm in solving complex high-dimensional optimization problems with a slow convergence speed, low accuracy, and the high likelihood to fall into local optimum, a mixed Harris hawks optimization (MHHO) algorithm based on the pinhole imaging strategy is proposed, including four strategies to improve the optimization performance. Firstly, the pinhole imaging strategy is used to enable the Harris' hawks to approach the optimal solution faster and accelerate convergence. Secondly, the spiral parameter is introduced into the exploration phase to help the searching paths of the Harris' hawks more diverse and improve the global search ability of the algorithm. Finally, the greedy strategy of the aquila optimization algorithm and the position update strategy of the flower pollination optimization algorithm are embedded in the exploitation stage to make the algorithm jump out of local optimum effectively. To verify the effectiveness of the proposed MHHO algorithm, it is compared with the classical HHO algorithm and 16 other state-of-the-art algorithms, and extensively tested on 23 well-known benchmark functions, the IEEE CEC2017 test sets, and three complex constrained engineering optimization problems. The test results show that MHHO achieves the top ranking on both the benchmark functions and the CEC2017 test sets, demonstrating its superior performance in terms of faster convergence speed and higher accuracy. [ABSTRACT FROM AUTHOR]

Medical images contain sensitive information that will be shared with healthcare workers, patients, and researchers for diagnostic and research purposes. However, the security of these images during transmission is extremely important, especially after the COVID-19 pandemic. In diagnosis, MRI and CT images are two common medical images, and the two medical images complement each other to a certain extent. The fusion of these two images can help doctors diagnose more intuitively. This paper proposes a zero-watermarking encryption scheme with an authentication function based on multimodal medical image fusion. In addition, an improved three-dimensional Chebyshev and ICMIC closed-loop coupling mapping (3D-CICCM) is proposed and applied to the design of medical image encryption. This scheme mainly includes: medical image fusion based on bilateral filtering and local gradient energy, authentication based on the Fresnel transform, double random phase encryption based on the Fresnel transform and shuffle algorithm, and zero-watermark generation algorithm based on stationary wavelet transform. The experimental results prove the effectiveness of this scheme. [ABSTRACT FROM AUTHOR]

Multi-layer perceptron (MLP) is a class of Artificial Neural Networks widely used in regression, classification, and prediction. To accelerate the training of MLP, more cores can be used for parallel computing on many-core systems. However, with the increasing number of cores integrated into the chip, the communication bottleneck in the training of MLP on electrical network-on-chip (ENoC) becomes severe, degrading MLP training performance. Replacing ENoC with optical network-on-chip (ONoC) can break the communication bottleneck in MLP training. To facilitate the development of ONoC for MLP training, it is necessary to compare and model the MLP training performance of ONoC and ENoC in advance. This paper first analyzes and compares the differences between ONoC and ENoC. Then, we formulate the performance and energy model of MLP training on ONoC and ENoC by analyzing the communication and computation time, static energy, and dynamic energy consumption, respectively. Furthermore, we conduct extensive simulations to compare their MLP training performance and energy consumption with our simulation infrastructure. The experimental results show the MLP training time of ONoC has been reduced by 65.16% and 52.51% on average in different numbers of cores and batch sizes compared with ENoC. The results also exhibit that ONoC overall has 54.86% and 43.13% on average energy reduction in different numbers of cores and batch sizes compared with ENoC. However, with a small number of cores (e.g., less than 50) in MLP training, ENoC consumes less energy than ONoC. These experiments confirm that generally ONoC is a good replacement for ENoC when using a large number of cores in terms of performance and energy consumption for MLP training. [ABSTRACT FROM AUTHOR]

The study is designed to improve the efficiency of Internet of Things (IoT) structure detection and achieve the smooth operation of IoT networks. First, the connection between the IoT network structure and maxdegree is investigated based on analyzing the IoT supertask network structure to find the main influence factor of maxdegree, along with the conditions for obtaining the optimal maxdegree. Second, a structural algorithm model of optimal supertask network is proposed as the foundation for achieving the minimum maxdegree. Finally, the human behavior recognition database is taken as the research object to verify its performance through the specific instance data. The IoT network structure factors are proved to include the task quantity, resource capacity, number of networks, and Communication Calculation Ratio (CCR). The experimental results also show that the principal factor that affects maxdegree is the number of different tasks. Besides, there is a mutually positive interaction between the network structure and the IoT maxdegree, which complement each other and form the core network of IoT. Moreover, the results reflect the good performance on different datasets of the supertask IoT network structure for the human behavior recognition database. There exists the optimal maxdegree of the model under the condition of 40 tasks, 32 resources, 6 networks and CCR of 6. Furthermore, the proposed algorithm has a shorter length and lower complexity than other related algorithms, which is very suitable for the construction of IoT networks. The research results can provide some references and practical value for the construction and data processing of the IoT structure. [ABSTRACT FROM AUTHOR]