PyFLANK, a graph neural network based null distribution inference model for FST outlier detection.

This study aimed to investigate the effectiveness and practicality of using models like convolutional neural network and transformer in detecting and precise segmenting meningioma from magnetic resonance images. The retrospective study on T1-weighted and contrast-enhanced images of 523 meningioma patients from 3 centers between 2010 and 2020. A total of 373 cases split 8:2 for training and validation. Three independent test sets were built based on the remaining 150 cases. Six convolutional neural network detection models trained via transfer learning were evaluated using 4 metrics and receiver operating characteristic analysis. Detected images were used for segmentation. Three segmentation models were trained for meningioma segmentation and were evaluated via 4 metrics. In 3 test sets, intraclass consistency values were used to evaluate the consistency of detection and segmentation models with manually annotated results from 3 different levels of radiologists. The average accuracies of the detection model in the 3 test sets were 97.3%, 93.5%, and 96.0%, respectively. The model of segmentation showed mean Dice similarity coefficient values of 0.884, 0.834, and 0.892, respectively. Intraclass consistency values showed that the results of detection and segmentation models were highly consistent with those of intermediate and senior radiologists and lowly consistent with those of junior radiologists. The proposed deep learning system exhibits advanced performance comparable with intermediate and senior radiologists in meningioma detection and segmentation. This system could potentially significantly improve the efficiency of the detection and segmentation of meningiomas.

Anomaly detection in time series data (e.g., sensor data) is becoming a fundamental research problem that has various applications. Due to the complex inter-sensor relationships, it is challenging to detect anomalous events such as system faults and attacks hidden the high-dimensional time series. Recent advancements in deep learning approaches such as Graph Neural Networks (GNN) have greatly improved anomaly detection performance in time series data. However, existing methods usually use separate components to capture spatial and temporal dependence relationship and ignore the heterogeneities in spatial-temporal data which motivates us to capture them together. In this paper, we propose a novel approach STGDNN (short for Spatial-Temporal Graph Deviation Neural Networks) that learns the spatial-temporal dependence relationship together in structure learning of graph neural networks. In addition, we use the learned spatial-temporal graph structure and attention weights to explain the detected anomalies. The experiments on three real world datasets show our superiority in detection accuracy, anomaly diagnosis, and model interpretation compared with state-of-the-art methods.

A recent paper (1) purported to document negligible climatic determination among European bird species, with implications for forecasting range shifts in changing climates. However, only 12 of 100 species analyzed were endemic—thus, for the remaining 88% of test species, key limits with likely climatic determination were excluded, particularly eastward and southward (2). Second, the authors developed null distributions conserving the same prevalence and semivariogram as real species' distributions and showed that real distributions were not modeled better than “null” distributions. However, most variation in null and real distributions at broad geographic scales is explained by spatially structured climate variation that is difficult to disentangle. The authors' manipulation fails to eliminate climate as a correlate of null ranges exactly because their occurrences were spatially clumped: as climate is autocorrelated, null distributions have climate signatures just like real distributions. Producing null models removing effects of climate but keeping the spatial cohesion of the distribution is unlikely, as in recent debates regarding mid-domain effects as appropriate null models for diversity (3). Null distributions will show climate signatures similar to those of real species that are not eliminated in their randomization algorithm, so the authors cannot reject the hypothesis of a climate association in European birds. Nonetheless, we agree that, because distributions can be modeled by using current spatially structured climatic predictors without necessity of direct causal linkages (4), more studies oriented at testing the robustness of correlative methods in predicting species' distributions under future climate scenarios are needed (5).

Recently, edge computing has emerged as a prevailing paradigm in applying deep learning-based object detection models, offering a promising solution for time-sensitive tasks. However, existing edge object detection faces several challenges: 1) These methods struggle to balance detection precision and model lightweightness. 2) Existing generalized edge-deployment designs offer limited adaptability for object detection. 3) Current works lack real-world evaluation and validation. To address these challenges, we propose the <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"><u>E</u>dge <u>D</u>etection <u>Toolbox</u></i> (ED-TOOLBOX), which leverages generalizable plug-and-play components to enable edge-site adaptation of object detection models. Specifically, we propose a lightweight <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Reparameterized Dynamic Convolutional Network</i> (Rep-DConvNet) that employs a weighted multi-shape convolutional branch structure to enhance detection performance. Furthermore, ED-TOOLBOX includes a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Sparse Cross-Attention</i> (SC-A) network that adopts a localized-mapping-assisted self-attention mechanism to facilitate a well-crafted <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Joint Module</i> in adaptively transferring features for further performance improvement. Moreover, we propose an <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Efficient Head</i> for the classification and location modules to achieve more efficient prediction. Additionally, in practical industrial scenarios, we identify that helmet detection-one of the most representative edge object detection tasks-overlooks band fastening, which introduces potential safety hazards. To address this, we build a <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Helmet Band Detection Dataset</i> (HBDD) and apply an edge object detection model optimized by the ED-TOOLBOX to tackle this real-world task. Extensive experiments validate the effectiveness of components in ED-TOOLBOX. In visual surveillance simulations, ED-TOOLBOX-assisted edge detection models outperform six <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">state-of-the-art</i> methods, enabling real-time and accurate detection. These results demonstrate that our approach offers a superior solution for edge object detection.

The increasing prevalence of cyber threats across Internet of Medical Things (IoMT) ecosystems poses critical challenges for safeguarding patient safety and data integrity, necessitating a dynamic, resilient intrusion detection system (IDS). In this work, we present a comprehensive machine learning framework for classifying cyberattacks in IoMT settings using biometric and network traffic data from the publicly available WUSTL-EHMS-2020 dataset. We conduct a unique comparative analysis using three paradigms: a Graph Neural Network (GNN) model to capture structural dependencies a Transformer deep learning model to capture contextual relationships and a lightweight baseline classifier, Logistic Regression. We undertook extensive data preparation, including label encoding, normalisation, and stratified sampling to maintain class balance. The Transformer achieved the highest overall classification accuracy in the IoMT ecosystem (93.5 parcent), outperforming both GNN (88.7 parcent) and Logistic Regression (92.8 parcent) across all evaluation metrics. Our research demonstrates the superior ability of attention-based models to identify complex threat patterns in heterogeneousIoMT data. Our study provides a reproducible benchmarking framework and lays the groundwork for future efforts related to hybrid modelling, explainable AI, and federated learning to improve the cybersecurity of Smart Healthcare Systems.

In many practical classification problems, datasets would have a portion of outliers, which could greatly affect the performance of the constructed models. In order to address this issue, we apply the group method of data handing (GMDH) neural network in outlier detection. This study builds a GMDH-based outlier detection (GOD) model. This model first implements feature selection in the training set L using GMDH neural network. Then a new training set L′ can be obtained by mapping the selected key feature subset. Next, a linear regression model can be constructed in the set L′ by ordinary least squares estimation. Further, it eliminates a sample from the set L′ randomly every time, and then rebuilds a linear regression model. Finally, outlier detection is realized by calculating Cook’s distance for each sample. Four different customer classification datasets are used to conduct experiments. Results show that GOD model can effectively eliminate outliers, and compared with the five existing outlier detection models, it generally performs significantly better. This indicates that eliminating outliers can effectively enhance classification accuracy of the trained classification model.

AI-enabled and multimodal data driven smart health monitoring of wind power systems: A case study

Anomaly detection in ad-hoc networks based on deep learning model: A plug and play device

The rise of social media platforms has led to an increase in the flow and dissemination of information, but it has also made generating and spreading rumors easier. Rumor detection requires understanding the context and semantics of text, dealing with the evolving nature of rumors, and processing vast amounts of data in real-time. Deep learning (DL)-based techniques exhibit a higher accuracy in detecting rumors on social media compared to many traditional machine learning approaches. This study presents a systematic review of DL approaches in rumor detection, analyzing datasets, pre-processing methods, feature taxonomy, and frequently used DL methods. In the context of feature selection, we categorize features into three areas: text-based, user-based, and propagation-based. Besides, we surveyed the trends in DL models for rumor detection and classified them into convolutional neural networks (CNN), recurrent neural networks (RNN), graph neural networks (GNN), and other methods based on the model structure. It offers insights into effective algorithms and strategies, aiming to guide researchers, developers, social media users, and governments in detecting and preventing the spread of false information. The study contributes to enhancing research in this field and identifies potential areas for future exploration.

Nowadays, the popularity of intelligent terminals makes malwares more and more serious. Among the many features of application, the call graph can accurately express the behavior of the application. The rapid development of graph neural network in recent years provides a new solution for the malicious analysis of application using call graphs as features. However, there are still problems such as low accuracy. This paper established a large-scale data set containing more than 40,000 samples and selected the class call graph, which was extracted from the application, as the feature and used the graph embedding combined with the deep neural network to detect the malware. The experimental results show that the accuracy of the detection model proposed in this paper is 97.7%; the precision is 96.6%; the recall is 96.8%; the F1-score is 96.4%, which is better than the existing detection model based on Markov chain and graph embedding detection model.

A large amount of work has been done on the KDD 99 dataset, most of which include the use of a hybrid anomaly and misuse detection model done in parallel with each other. In order to further classify the intrusions, our approach to network intrusion detection includes the use of two different anomaly detection models followed by misuse detection applied to the combined output obtained from the previous step. The end goal of this is to verify the anomalies detected by the anomaly detection algorithm and clarify whether they are actually intrusions or random outliers from the trained normal (and thus to try and reduce the number of false positives). We aim to detect a pattern in this novel intrusion technique itself, and not the handling of such intrusions. The intrusions were detected to a very high degree of accuracy.

The rate of non-medical use of opioid drugs has increased markedly since the early 2000s. Due to this non-medical use, abusers suffer from different adverse effects that include physical and psychological problems. Many studies have been done to detect Drug Abuse (DA) events from social media data using machine learning and deep learning concepts. Moreover, Graph Neural Networks (GNNs) have recently become popular in text classification tasks due to their high accuracy and capability to handle complex structures. In this work, we collect drugs-related Twitter data (tweets) and build text graphs (corpus-level and document-level) to capture word-word, document-word, and document-document relations. Then we apply different GNN models on those text graphs and thus turn the text classification task into a node classification (for corpus-level graph) and graph classification (for document-level graph) task to detect DA events. Finally, we compare our graph-based DA detection models with different types of baselines models, including rule-based, traditional machine learning, and deep learning models. Our result shows graph-based models outperform the traditional machine learning and deep learning-based models.

In the "blind review" mode for technical bid evaluation within the power industry, review experts must make impartial judgments based on anonymized technical response documents. However, these documents often contain a large amount of complex graphical content, which existing general Optical Character Recognition (OCR) technology struggles to accurately identify. To address this, this study first constructs a specialized image dataset covering typical power industry response document scenarios. Subsequently, an improved You Only Look Once (YOLO) object detection model is developed, incorporating an anchor box clustering strategy optimized for small, densely packed power symbols and a multi-scale feature fusion mechanism. Then, a local contextual relationship inference module based on Graph Neural Networks is designed and integrated. Finally, an end-to-end recognition pipeline from image input to structured information output is constructed. Experimental results indicate that the improved model achieves a mean Average Precision (mAP) of 90.0% in key equipment detection tasks; in topology relationship restoration experiments, the average Graph Edit Distance (GED) is 0.172; in end-to-end efficiency experiments simulating real blind review scenarios, the average processing time per document is only 8.7 seconds, with an information completeness rate of 98.2%. The research demonstrates that the YOLO-based object detection and relationship modeling method, customized for the blind review mode, can effectively address the challenge of recognizing complex image content in power response documents, providing reliable technical support for the efficiency and fairness of blind reviews.

The financial property of Ethereum makes smart contract attacks frequently bring about tremendous economic loss. Method for effective detection of vulnerabilities in contracts imperative. Existing efforts for contract security analysis heavily rely on rigid rules defined by experts, which are labor-intensive and non-scalable. There is still a lack of effort that considers combining expert-defined security patterns with deep learning. This paper proposes EtherGIS, a vulnerability detection framework that utilizes graph neural networks (GNN) and expert knowledge to extract the graph feature from smart contract control flow graphs (CFG). To gain multi-dimensional contract information and reinforce the attention of vulnerability-related graph features, sensitive EVM instruction corpora are constructed by analyzing EVM underlying logic and diverse vulnerability triggering mechanisms. The characteristic of nodes and edges in a CFG is initially confirmed according to the corpora, generating the corresponding attribute graph. GNN is adopted to aggregate the whole graph's attribute and structure information, bridging the semantic gap between low-level graph features and high-level contract features. The feature representation of the graph is finally input into the graph classification model for vulnerability detection. Furthermore, automated machine learning (AutoML) is adopted to automate the entire deep learning process. Data for this research was collected from Ethereum to build up a dataset of six vulnerabilities for evaluation. Experimental results demonstrate that EtherGIS can productively detect vulnerabilities in Ethereum smart contracts in terms of accuracy, precision, recall, and F1-score. All aspects outperform the existing work.

Early detection of autism spectrum disorder (ASD) is particularly important given its insidious qualities and the high cost of the diagnostic process. Currently, static functional connectivity studies have achieved significant results in the field of ASD detection. However, with the deepening of clinical research, more and more evidence suggests that dynamic functional connectivity analysis can more comprehensively reveal the complex and variable characteristics of brain networks and their underlying mechanisms, thus providing more solid scientific support for computer-aided diagnosis of ASD. To overcome the lack of time-scale information in static functional connectivity analysis, in this paper, we proposes an innovative GNN-LSTM model, which combines the advantages of long short-term memory (LSTM) and graph neural networks (GNNs). The model captures the spatial features in fMRI data by GNN and aggregates the temporal information of dynamic functional connectivity using LSTM to generate a more comprehensive spatio-temporal feature representation of fMRI data. Further, a dynamic graph pooling method is proposed to extract the final node representations from the dynamic graph representations for classification tasks. To address the variable dependence of dynamic feature connectivity on time scales, the model introduces a jump connection mechanism to enhance information extraction between internal units and capture features at different time scales. The model achieves remarkable results on the ABIDE dataset, with accuracies of 80.4% on the ABIDE I and 79.63% on the ABIDE II, which strongly demonstrates the effectiveness and potential of the model for ASD detection. This study not only provides new perspectives and methods for computer-aided diagnosis of ASD but also provides useful references for research in related fields.