Graph neural network‐enhanced auxiliary classifier generative adversarial network framework for robust intrusion detection

TRA-ACGAN: A motor bearing fault diagnosis model based on an auxiliary classifier generative adversarial network and transformer network

Graph neural networks (GNNs) have proven their efficacy in a variety of real-world applications, but their underlying mechanisms remain a mystery. To address this challenge and enable reliable decision-making, many GNN explainers have been proposed in recent years. However, these methods often encounter limitations, including their dependence on specific instances, lack of generalizability to unseen graphs, producing potentially invalid explanations, and yielding inadequate fidelity. To overcome these limitations, we, in this paper, introduce the Auxiliary Classifier Generative Adversarial Network (ACGAN) into the field of GNN explanation and propose a new GNN explainer dubbed ACGAN-GNNExplainer. Our approach leverages a generator to produce explanations for the original input graphs while incorporating a discriminator to oversee the generation process, ensuring explanation fidelity and improving accuracy. Experimental evaluations conducted on both synthetic and real-world graph datasets demonstrate the superiority of our proposed method compared to other existing GNN explainers.

Convolutional neural networks (CNNs) have been widely used for gas‐insulated switchgear (GIS) partial discharge (PD) pattern recognition due to their powerful feature extraction ability. However, there is commonly a scarcity of fault samples due to low insulation failure rate of GIS equipment, which degrades the diagnostic performance of these CNN networks when directly applied to small and unbalanced datasets. Therefore, we propose a novel auxiliary classifier generative adversarial network for GIS PD pattern recognition for small and unbalanced samples. First, we propose using synchrosqueezed wavelet transform to extract time‐frequency characteristics of PD pulses and obtain a time‐frequency image with high energy aggregation and time‐frequency distribution rate. Then, we propose an improved generative adversarial network with an auxiliary classier and self‐attention mechanism, which can generate high‐quality PD samples for situations with few classes. Experiments show that our proposed method can reach 95.75% recognition accuracy for small datasets, which is the highest among several comparable methods. Furthermore, the proposed method has excellent and stable recognition performance for various unbalanced datasets.

Deep-learning-based SAR ship classification has become a research hotspot in the military and civilian fields and achieved remarkable performance. However, the volume of available SAR ship classification data is relatively small, meaning that previous deep-learning-based methods have usually struggled with overfitting problems. Moreover, due to the limitation of the SAR imaging mechanism, the large intraclass diversity and small interclass similarity further degrade the classification performance. To address these issues, we propose a label smoothing auxiliary classifier generative adversarial network with triplet loss (LST-ACGAN) for SAR ship classification. In our method, an ACGAN is introduced to generate SAR ship samples with category labels. To address the model collapse problem in the ACGAN, the smooth category labels are assigned to generated samples. Moreover, triplet loss is integrated into the ACGAN for discriminative feature learning to enhance the margin of different classes. Extensive experiments on the OpenSARShip dataset demonstrate the superior performance of our method compared to the previous methods.

Covert timing channels (CTCs) are defined as a mechanism that embeds covert information into network traffic. In a manner, information leakage caused by CTCs brings serious threat to network security. In recent years, detection of CTCs is a focus and a challenging task in the field of covert channel research. However, existing detection schemes based on statistical methods have poor performance in detecting multiple CTCs, and require so many inter-arrival times of packets that these schemes cannot detect CTCs in real time. In this paper, we propose a novel deep learning approach for CTCs detection, namely, covert timing channels detection based on auxiliary classifier generative adversarial network (CD-ACGAN). The network structure and loss function of CD-ACGAN are designed to be suitable for CTCs detection task. We first encode traffic flows into single-channel Gramian Angular Field (GAF) images. Then we use CD-ACGAN to learn features from GAF images and predict the classes of CTCs. Our experimental results show that our approach has high accuracy and strong robustness in detecting various CTCs.

Effective response strategies to earthquake disasters are crucial for disaster management in smart cities. However, in regions where earthquakes do not occur frequently, model construction may be difficult due to a lack of training data. To address this issue, there is a need for technology that can generate earthquake scenarios for response training at any location. We proposed a model for generating earthquake scenarios using an auxiliary classifier Generative Adversarial Network (AC-GAN)-based data synthesis. The proposed ACGAN model generates various earthquake scenarios by incorporating an auxiliary classifier learning process into the discriminator of GAN. Our results at borehole sensors showed that the seismic data generated by the proposed model had similar characteristics to actual data. To further validate our results, we compared the generated IM (such as PGA, PGV, and SA) with Ground Motion Prediction Equations (GMPE). Furthermore, we evaluated the potential of using the generated scenarios for earthquake early warning training. The proposed model and algorithm have significant potential in advancing seismic analysis and detection management systems, and also contribute to disaster management.

With the growing popularity of smartphones, continuous and implicit authentication of such devices via behavioral biometrics such as touch dynamics becomes an attractive option, especially when the physical biometrics are challenging to utilize, or their frequent and continuous usage annoys the user. However, touch dynamics is vulnerable to potential security attacks such as shoulder surfing, camera attack, and smudge attack. As a result, it is challenging to rule out genuine imposters while only relying on models that learn from real touchstrokes. In this paper, a touchstroke authentication model based on Auxiliary Classifier Generative Adversarial Network (AC-GAN) is presented. Given a small subset of a legitimate user's touchstrokes data during training, the presented AC-GAN model learns to generate a vast amount of synthetic touchstrokes that closely approximate the real touchstrokes, simulating imposter behavior, and then uses both generated and real touchstrokes in discriminating real user from the imposters. The presented network is trained on the Touchanalytics dataset and the discriminability is evaluated with popular performance metrics and loss functions. The evaluation results suggest that it is possible to achieve comparable authentication accuracies with Equal Error Rate ranging from 2% to 11% even when the generative model is challenged with a vast number of synthetic data that effectively simulates an imposter behavior. The use of AC-GAN also diversifies generated samples and stabilizes training.

The inherent complexity and heterogeneity of the Internet of Things (IoT) ecosystem present significant challenges for developing effective intrusion detection systems. While graph deep-learning-based methods have shown promise in cybersecurity applications, existing approaches primarily construct graphs based on physical network connections, which may not effectively capture node representations. This paper proposes a Top-K Similarity Graph Framework (TKSGF) for IoT network intrusion detection. Instead of relying on physical links, the TKSGF constructs graphs based on Top-K attribute similarity, ensuring a more meaningful representation of node relationships. We employ GraphSAGE as the Graph Neural Network (GNN) model to effectively capture node representations while maintaining scalability. Furthermore, we conducted extensive experiments to analyze the impact of graph directionality (directed vs. undirected), different K values, and various GNN architectures and configurations on detection performance. Evaluations on binary and multi-class classification tasks using the NF-ToN IoT and NF-BoT IoT datasets from the Machine-Learning-Based Network Intrusion Detection System (NIDS) benchmark demonstrated that our proposed framework consistently outperformed traditional machine learning methods and existing graph-based approaches, achieving superior classification accuracy and robustness.

Summary Deep learning have led to a certain extent breakthrough in seismic facies analysis(SFA). The greatest challenge in the SFA domain is how to cope with large seismic datasets and the limited amount of annotated samples, especially when employing supervised machine learning algorithms that require labelled data and larger training sample. we develop a supervised seismic facies analysis via an improved Auxiliary Classifier Generative Adversarial Network(ACGAN).To solve the problem of imbalanced seismic facies class distribution, and limited training data with labels, we used an improved ACGAN to generate synthetic seismic data sets, and then train Convolutional Neural Networks(CNN) using theses generated synthetic data. Fault data and seismic facies data test demonstrated that generated seismic data via the improved ACGAN can be used for synthetic data augmentation as for increased the performance of CNN and avoided overfitting compared to original seismic data training set. Therefore, we believe that the proposed method can effectively alleviate the problems of insufficient and imbalanced seismic data in SFA, and also can generalize to other seismic data classification applications.

Deep neural networks require a large amount of data for supervised training and learning, but it is often difficult to obtain a large amount of label data in practical applications. Since semi-supervised learning can reduce the dependence of deep networks on label data, the generative confrontation network based on semi-supervised learning can improve the classification effect. Although researchers have made progress in semi-supervised learning, small-scale, fully-supervised tasks have not been solved well, because even unlabeled data cannot be obtained in large quantities in such tasks. Therefore, we propose a new GAN model EAC-GAN based on the auxiliary classifier generative adversarial network (ACGAN), which uses genetic algorithms and semi-supervised algorithms to improve classification under fully supervised conditions. Our method utilizes ACGAN to generate artificial data for supplementing supervised classification. More specifically, we attached an external classifier to the original ACGAN generator, so it was named EAC-GAN instead of sharing an architecture with the discriminator. Our experiments show that the performance of EAC-GAN is far superior to standard methods based on data augmentation and regularization, and it is effective on a small amount of real data sets.KeywordsImage synthesisData enhancementGenerative adversarial networkSemi-supervised learning

Accurate classification of brain tumors is often limited by the scarcity of labeled medical images, particularly for complex tumor subtypes. This study proposes an auxiliary classifier generative adversarial network (ACGAN) to generate class-conditioned synthetic MRI images and to enhance the performance of tumor-classification models. The ACGAN was trained using the publicly available Figshare Brain MRI dataset, which contains four categories: glioma, meningioma, pituitary tumor, and healthy tissue. After training, the model produced high-fidelity synthetic MRI images that were quantitatively evaluated using the Fréchet inception distance (FID) as a distribution-level proxy metric, achieving a score of 4.6 and indicating close alignment between the real and synthetic data distributions. In addition to generative quality, the classification network trained with ACGAN-augmented data achieved 94.8% accuracy, 95.3% sensitivity, 93.6% specificity, and 94.1% F1-score, outperforming models trained solely on real data. Comparative analysis demonstrates that the proposed framework provides superior class separability and lower FID values than commonly used generative methods, including DCGAN and standard cGAN approaches. These findings highlight the potential of ACGAN-based synthetic augmentation to support medical-imaging tasks under limited-data conditions and to improve the reliability of brain-tumor classification systems.

B-mode ultrasound imaging is a popular medical imaging technique. Like other image processing tasks, deep learning has been used for analysis of B-mode ultrasound images in the last few years. However, training deep learning models require large labeled datasets, which is often unavailable for ultrasound images. The lack of large labeled data is a bottleneck for the use of deep learning in ultrasound image analysis. To overcome this challenge, in this work, we exploit auxiliary classifier generative adversarial network (ACGAN) that combines the benefits of data augmentation and transfer learning in the same framework. We conduct experiment on a dataset of breast ultrasound images that shows the effectiveness of the proposed approach.

Expressiveness is a critical feature for the communication between humans and robots, and it helps humans to better understand and accept a robot. Emotions can be expressed through a variety of modalities: kinesthetic (via facial expression), body posture and gestures, auditory, thus the acoustic features of speech, and semantic, thus the content of what is said. One of the most effective modalities to communicate emotions is through facial expressions. Social robots often show facial expressions with coded animations. However, the robot must be able to express appropriate emotional responses according to the interaction with people. In this work, we consider verbal interactions between humans and robots and propose a system composed of two modules for the generation of facial emotions by recognising the arousal and valence values of a written sentence. The first module, based on Bidirectional Encoder Representations from Transformers, is deployed for emotion recognition in a sentence. The second, an Auxiliary Classifier Generative Adversarial Network, is proposed for the generation of facial movements for expressing the recognised emotion in terms of valence and arousal.

Deep neural networks (DNNs) are increasingly deployed in various fields. Despite their spectacular advances, DNNs are known to suffer from adversarial vulnerabilities. The robustness of DNNs is then threatened by leading them to misclassifications with unexpected inputs (adversarial examples). The fuzzing technique frequently used for testing traditional software has recently been adopted to evaluate the robustness of DNNs. Current DNN fuzzing techniques focus on image classification DNNs and generate test cases by mutations, e.g., image transformations and adversarial perturbations. However, mutation-based test cases usually lack diversity and have distribution deflection from the original DNN input space, which impacts the evaluation of DNNs. In this paper, we propose a generation-based fuzzing frame-work FuzzGAN to detect adversarial flaws existing in DNNs. We integrate the testing purpose and the guidance of the neuron coverage into the original objectives of auxiliary classifier generative adversarial networks. Hence, FuzzGAN learns the representation of a DNN's input space and generates test cases without the limitation of any concrete seed input. We evaluate the performance of FuzzGAN on two DNN models that have classical network structures and are trained on public datasets. The experiment results demonstrate that FuzzGAN can generate realistic, diverse and valid test cases and achieve high neuron coverage. Moreover, these test cases can be used to improve the performance of the target DNN through adversarial retraining.

Electroencephalogram (EEG) signal is often used to assess sleep quality and treat sleep disorders. Many existing methods usually obtain high accuracy through a large number of feature preprocessing and feature extraction of EEG signals, which need a lot of prior knowledge as the basis. In this paper, a novel sleep stage classification framework, named FRL&S2SL, is proposed. The framework combines fast representation learning (FRL) and semantic-to-signal learning (S2SL) and uses single-channel EEG without any preprocessing of EEG signals. In the proposed framework, we utilize convolutional neural networks (CNN) to extract time-invariant features and bidirectional long short-term memory (BiLSTM) models to extract temporal features. Furthermore, auxiliary classifier generative adversarial network (ACGAN) is used to embed semantic features into signal features and to extract knowledge domain features of EEG signals for the first time. According to the American Academy of Sleep Medicine (AASM), sleep is divided into five stages: awake, rapid eye movement (REM) and three non-rapid eye movement (N1/N2/N3). We evaluated our framework using single-channel EEG (Fpz-Oz) from Sleep-EDF dataset, which is subject to the standards specified by AASM. The results show that our framework has achieved state-of-the-art in many evaluation metrics.