Graph embedding is the procedure of transforming a Graph into a low-dimensional, informative representation. The majority of existing Graph embedding techniques have given less consideration to the embedding distribution of the latent codes and more attention to the Graph’s structure. Recently, Variational Graph Autoencoders (VGAEs) have demonstrated good performance by learning smooth representations from unlabeled training samples. On the other hand, in regular VGAEs, the prior distribution over latent variables is generally a single Gaussian distribution. However, complex data distributions cannot be well-modelled under the assumption of a single Gaussian distribution. This choice of prior distribution is important because each dimension of a multivariate Gaussian can learn a separate continuous latent feature, which can result more structured and disentangled representation. In this paper, we employ the Gaussian Mixture Model (GMM) as the prior distribution in a Variational Graph Autoencoder (GMM-VGAE) framework for node classification in Graphs. In this framework, GMM effectively discovers the inherent complex data distribution, and Graph convolutional networks (GCNs) exploit the structure of the nodes of a Graph to learn more informative representations. The proposed model incorporates several Graph Convolutional Networks (GCNs): one to map the input feature vector to the latent representation utilized for classification, another to generate the parameters of the latent distribution for learning from unlabeled data, and finally, an additional GCN is employed for reconstructing the input and delivering the reconstruction loss. Through extensive experiments on well-known Citations, Co-authorship, and Social network Graphs, GMM-VGAE’s superiority over state-of-the-art methods is demonstrated.

Purpose: The objective of this paper is to study the feasibility of using effective connectivity (Granger Causality) (GC) obtained from resting-state functional Magnetic Resonance Imaging (rs-fMRI) data and Stacked Autoencoder for diagnosing Autism Spectrum Disorder (ASD) and comparing the results with those obtained using functional connectivity (Pearson Correlation Coefficient) (PCC). ASD affects the normal development of the brain in the field of social interactions and communication skills. Because diagnosing ASD using behavioral symptoms is a time-consuming subjective process that needs the exact collaboration of the ASD subject or his/her relatives, in recent years diagnosing ASD using resting-state functional neuroimaging modalities like rs-fMRI, has been taken into consideration. Materials and Methods: We used rs-fMRI data and compared the use of functional and effective connectivity features using an Autoencoder to classify people with ASD from healthy subjects. We used ABIDE dataset and divided the brain into 100 regions using the Harvard-Oxford (HO) Atlas. We calculated the PCC in classification using functional connectivity, and we calculated the GC in classification using effective connectivity. We used a Stacked Autoencoder to reduce the dimension of feature-space and a multi-layered perceptron (MLP) neural network as a classifier in both classifications. Results: We achieved an accuracy of 67.8%, a sensitivity of 68.5%, and a specificity of 66.6% in classification using functional connectivity, and we achieved an accuracy of 67.6%, a sensitivity of 73.1%, and a specificity of 60.8% in classification using effective connectivity.  Conclusion: Although the accuracy obtained using functional and effective connectivity are almost similar, the sensitivity is notably higher using effective connectivity. Since sensitivity is more important than specificity in the medical diagnosis, it seems that using effective connectivity features may outperform the ASD diagnosis in practice. The purpose of this paper is to diagnose ASD using effective connectivity measures and deep neural network by rs-fMRI data, but we compare its results with functional connectivity measures. As far as we know, this is the first time that Granger Causality (GC) and Stacked Autoencoder have been used to diagnose ASD together.

With the widespread use of Android smartphones, the Android platform has become an attractive target for cybersecurity attackers and malware authors. Meanwhile, the growing emergence of zero-day malware has long been a major concern for cybersecurity researchers. This is because malware that has not been seen before often exhibits new or unknown behaviors, and there is no documented defense against it. In recent years, deep learning has become the dominant machine learning technique for malware detection and could achieve outstanding achievements. Currently, most deep malware detection techniques are supervised in nature and require training on large datasets of benign and malicious samples. However, supervised techniques usually do not perform well against zero-day malware. Semi-supervised and unsupervised deep malware detection techniques have more potential to detect previously unseen malware. In this paper, we present MalGAE, a novel end-to-end deep malware detection technique that leverages one-class Graph neural networks to detect Android malware in a semi-supervised manner. MalGAE represents each Android application with an attributed function call Graph (AFCG) to benefit the ability of Graphs to model complex relationships between data. It builds a deep one-class classifier by training a Stacked Graph Autoencoder with Graph convolutional layers on benign AFCGs. Experimental results show that MalGAE can achieve good detection performance in terms of different evaluation measures.

Purpose: Sleep is a subconscious state, and the brain is active during it. Automatic classification of sleep stages can help identify various diseases. In recent years, automatic sleep monitoring using deep learning networks has attracted the attention of researchers. Materials and Methods: In this paper, a deep learning type neural network called Stacked Autoencoders (SAEs) is used for automatically classifying sleep stages. SAEs are a kind of neural network with encoder and decoder blocks. The function of these networks is similar to the human brain and is capable of automatically processing signals,also SAEs are robust to noise. To prove the efficiency of this network, in addition to examining the effect of various biological signals such as Electrocardiogram (ECG) and Electroencephalogram (EEG) on the performance of sleep stage classification, Sleep Heart Health Study (SHHS) and ISRUC standard databases have been used, which include night recordings of 30 and 10 healthy humans, respectively. Results: The accuracy of classifying 2 to 6 classes by SHHS database are 0. 995, 0. 983, 0. 9780, 0. 9688, 0. 961, and on ISRUC database accuracies are 0. 996, 0. 994, 0. 9511, and 0. 9431. Moreover, the proposed network can classify wake, deep sleep, and light sleep using the ECG signal (acc = 0. 75, kappa = 0. 69). Conclusion: In the review of the results, it is concluded that sleep stages classification based on EEG signal has better results, still acquisition of ECG signal and its acceptable results can be a good alternative to use. In addition to its high ability of the proposed method to detect sleep stages, this network is robust to noise, which is very necessary and important for the clinical processing of sleep signals.

Crimes nowadays pose unique issues to security and legal institutions and requires smart approaches to different types of peculiar behavior within. This paper proposes a deep learning autoencodes framework to analyze and recognize unusual activities in the FBI’s crime dataset. Utilizing the Autoencoder model’s architecture consisting of input, compression, and output layers, the Adam optimizer is used with a Mean Squared Error loss function for training, validating with twenty percent of the data. A reconstruction error is calculated and subsequently, a threshold of the 95th percentile of the average MSE is set to flag anomalies. Findings prove that the model outperforms all comparative methodologies, achieving 98% accuracy and a 97% precision F1 score. In addition, the model was shown to have an AUC on ROC curve of 98.2% which confirms the model’s ability to accurately classify normal and abnormal samples. This study illustrates the capability of multi-dimensional Autoencoders to analyze and process complex crime data which can greatly aid security agencies in premeditative and reactive responses to crime. Further research will focus on attention-based hybrid models along with system for real-time responsive tracing of volatile hyperdynamics.