12 December 2019 in IEEE Access
Network intrusion detection systems (NIDS) are essential tools in ensuring network information security, and neural networks have become an increasingly popular solution for NIDS. However, with the gradual complexity of the network environment, the existing solutions using the conventional neural network cannot make full use of the rich information in the network traffic data due to its single structure. More importantly, this will lead to the existing NIDS have incomplete knowledge of the intrusion detection domain, and making it unable to achieve a high detection rate and good stability in the new environment. In this paper, we take a step forward and extract the different level features from the network connection, rather than a long feature vector used in the traditional approach, which can process feature information separately more efficiently. And further, we propose multimodal-sequential intrusion detection approach with special structure of hierarchical progressive network, which is supported by multimodal deep auto encoder (MDAE) and LSTM technologies. By design the special structure of hierarchical progressive network, our approach can efficiently integrate the different level features information within a network connection and automatically learn temporal information between adjacent network connections at the same time. Based on the three benchmark datasets from 1999 to 2017, including NSL-KDD, UNSW-NB15, and CICIDS 2017, we investigated the performance of our proposed approach on the task of detecting attacks within modern network. The experimental results show that the average accuracy of this method is 94% in binary classification and 88% in multi-class classification, which is at least 2% and 4% super than other methods respectively, and demonstrated that our model has excellent stability. Moreover, we further explore the multimodality and complementarity in traffic data, the experimental results show that the performance of detection model can be further improved in the range 2% to 5% when using our MDAE model to process the features of traffic data.
9 January 2020 in IEEE Access
With the rapidly increasing complexity and indispensable status of software systems, unprecedented challenges have been brought to software debugging and fault repair. Among the states of art automated fault localization techniques, spectrum-based fault localization (SBFL) is one of the most widely studied heuristic approaches. While, existing SBFL techniques are mostly focused on the analysis of the test spectrum, which loses sight of the effectiveness implied in interactions of software entities. In this paper, an optimized fault localization approach is proposed, which integrates spectrum-based fault localization and fault influence propagation analysis. The intuition is that the entity associated with more suspicious entities is more likely to be faulty. (1) A dynamic-instrumenting execution tracer is developed to record test coverage data and method call relations simultaneously. (2) The Fault Influence Network (FIN) based on complex network theory is constructed, of which the network topology is abstracted from method call relations and the weights of nodes are calculated applying a raw fault locator. Thus, the test spectrum and the method interactive relations are combined for further collaborative analysis. (3) A PageRank-based fault localization approach is proposed. By the computation of fault influence propagation, the suspiciousness score PR-Susp of the real faulty method can be enhanced significantly. Also, a candidate set pruning based on failing tests is implemented, which is used for the narrowing of investigation. Experiments are conducted over real-world fault dataset Defects4J with 33 raw spectrum-based fault locators, which proves that the proposed approach improves the baseline 14.9% averagely on the metric acc@5 with a growth rate of over 39% on all metrics of acc@1, acc@3, acc@5, MAP, and MWE.
1 February 2022 in ACS Omega
In the process of coal mining, a certain amount of gas will be produced. Environmental perception is very important to realize intelligent and unmanned coal mine production and operation and to reduce the accident rate of gas explosion and other disasters. The identification of geometric features of the coal mine working face is the main part of the environmental perception of the working face. In this study, we identify geometric features in a large-scale coal mine working face point cloud (we take the ball as an example) so as to provide a method for the environmental perception of the coal mine working face. On the basis of previous research, we upgrade the dynamic graph convolution neural network (DGCNN) for directly processing point clouds from two aspects: extracting local features and global features of point clouds. First, a multiscale dynamic graph convolution neural network (MS-DGCNN) is proposed, and the combination of max-pooling and average-pooling is used as the symmetry function. Second, we use MS-DGCNN to learn the features of a variety of geometric point clouds in the point cloud data set we make and then look for the ball in the large-scale point cloud of the coal mining working face. Finally, we compare the performance of MS-DGCNN with that of other deep neural networks directly processing point clouds. This study enables MS-DGCNN to obtain more powerful feature expression ability and enhance the generalization of the model. In addition, this study provides a solid foundation for the geometric feature identification of MS-DGCNN in the environmental perception of the coal mine working face and creates a precedent for the application of MS-DGCNN in the field of energy. At the same time, this study makes a beneficial exploration for the development of a transparent coal mine working face.
1 March 2022 in Chinese Journal of Electronics
2 December 2022 in Human-Centric Intelligent Systems
Few-shot relation extraction is one of the current research focuses. The key to this research is to fully extract the relation semantic information through very little training data. Intuitively, raising the relation semantics awareness in sentences can improve the efficiency of the model to extract relation features to alleviate the overfitting problem in few-shot learning. Therefore, we propose an enhanced relation semantic feature model based on prototype network to extract relations from few-shot texts. Firstly, we design a multi-level embedding encoder with position information and Transformer, which uses local information in the text to enhance the relation semantics representation. Secondly, the encoded relation features are fed into the novel prototype network, which designs a method that utilizes query prototype-level attention to guide the extraction of supporting prototypes, thereby enhancing the prototypes representation to better classify the relations in query sentences. Finally, through experimental comparison and discussion, we prove and analyze the effectiveness of the proposed multi-level embedding encoder, and prototype-level attention can enhance the stability of the model. Furthermore, our model has substantial improvements over baseline methods.
28 June 2024 in IEEE Transactions on Instrumentation and Measurement
Sleep apnea is a potentially serious yet often unnoticed sleep disorder characterized by repeated cessation and resumption of breathing. The non-invasive and real-time detection of sleep apnea is paramount in developing suitable therapeutic strategies. Recently, deep learning-based detection methods using Electrocardiography (ECG) have demonstrated considerable potential and convenience compared to traditional expert-reliant methods. However, challenges persist: extracting multi-scale sleep apnea-related features effectively from the ECG segment to be diagnosed, capturing varying-length transition features from the surrounding segments of the target one, and adaptively fusing the features of varying-length input segments. In this paper, we propose a Dual-MultiScale Interactive Attention-based Convolution Neural Network (DM-IACNN) for automatic sleep apnea detection using single-lead ECG signals. It composes a CNN-based interactive multi-scale extraction module to learn the feature maps from ECG intra-segment. Furthermore, we propose a temporal-wise attention module to enhance the modeling of temporal features derived from the interactive multi-scale extraction module. Finally, the proposed model accepts three segments of varying lengths obtained from adjacent ECG segments surrounding the target segment as multi-scale inputs. It features a scale-wise attention module to fuse the temporal feature maps of these three segments, capturing the multi-scale transition features between them. A series of experiments were conducted on the Apnea-ECG dataset to investigate the effectiveness of the proposed DM-IACNN model. The results demonstrate that the DM-IACNN model delivers robust performance, with an accuracy of 91.02% and 100% of per-segment and per-recording evaluation paradigm, which outperforms the current state-of-the-art methodologies.