Recurrent Graph Research Articles

Forecasting Epidemic Spread with Recurrent Graph Gate Fusion Transformers.

Predicting the unprecedented, nonlinear nature of COVID-19 presents a significant public health challenge. Recent advances in deep learning, such as Graph Neural Networks, Recurrent Neural Networks (RNNs), and Transformers, have enhanced predictions by modeling regional interactions, managing autoregressive time series, and identifying long-term dependencies. However, prior works often feature shallow integration of these models, leading to simplistic graph embeddings and inadequate analysis across different graph types. Additionally, excessive reliance on historical COVID-19 data limits the potential of utilizing time-lagged data, such as intervention policy information. To address these challenges, we introduce ReGraFT, a novel Sequence-to-Sequence model designed for robust long-term forecasting of COVID-19. ReGraFT integrates Multigraph-Gated Recurrent Units (MGRUs) with adaptive graphs, leveraging data from individual states, including infection rates, policy changes, and interstate travel. First, ReGraFT employs adaptive MGRU cells within an RNN framework to capture inter-regional dependencies, dynamically modeling complex transmission dynamics. Second, the model features a Self-Normalizing Priming layer using SELUs to enhance stability and accuracy across short, medium, and long-term forecasts. Lastly, ReGraFT systematically compares and integrates various graph types derived from fully connected layers, pooling, and attention-based mechanisms to provide a nuanced representation of inter-regional relationships. By incorporating lagged COVID-19 policy data, ReGraFT refines forecasts, demonstrating RMSE reductions of 2.39-35.92% compared to state-of-the-art models. This work provides accurate long-term predictions, aiding in better public health decisions. Our code is available at https://github.com/mfriendly/ReGraFT.

Physics-Embedded Recurrent Graph Neural Network for Fault Diagnosis of Complex Systems

Physics-Embedded Recurrent Graph Neural Network for Fault Diagnosis of Complex Systems

Robust recurrent graph convolutional network approach based sequential prediction of illicit transactions in cryptocurrencies

Money laundering has urged the need for machine learning algorithms for combating illicit services in the blockchain of cryptocurrencies due to its increasing complexity. Recent studies have revealed promising results using supervised learning methods in classifying illicit Bitcoin transactions of Elliptic data, one of the largest labelled data of Bitcoin transaction graphs. Nonetheless, all learning algorithms have failed to capture the dark market shutdown event that occurred in this data using its original features. This paper proposes a novel method named recurrent graph neural network model that extracts the temporal and graph topology of Bitcoin data to perform node classification as licit/illicit transactions. The proposed model performs sequential predictions that rely on recent labelled transactions designated by antecedent neighbouring features. Our main finding is that the proposed model against various models on Elliptic data has achieved state-of-the-art with accuracy and f1\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$f_1$$\\end{document}-score of 98.99% and 91.75%, respectively. Moreover, we visualise a snapshot of a Bitcoin transaction graph of Elliptic data to perform a case study using a backward reasoning process. The latter highlights the effectiveness of the proposed model from the explainability perspective. Sequential prediction leverages the dynamicity of the graph network in Elliptic data.

SEISMICITY of ARMENIA and ADJACENT TERRITORIES in 20182019

Here is the overview of the seismicity of the territory of the Republic of Armenia and surrounding countries for 2018 and 2019. More than 4000 local earthquakes have been processed. N=1417 earthquakes for 2018 and N=1416 earthquakes for 2019 were included in the catalogue of the earthquakes of the Republic of Armenia. Within the Republic the highest level of seismic activity is fixed in the north, and in the source zone of the Spitak earthquake 1988, MLV=6.9, I0=10. Analysis of earthquake recurrence graphs for 2018, 2019 showed that the slope of the earthquake recurrence graphs (2018=0.47 and 2019=0.42 respectively) is slightly lower in absolute value than its average value for 1993–2017 (=0.49). Earthquakes felt on the territory of Armenia for 2018 and 2019 are described.

Multi-organ spatiotemporal information aware model for sepsis mortality prediction

BackgroundSepsis is a syndrome involving multi-organ dysfunction, and the mortality in sepsis patients correlates with the number of lesioned organs. Precise prognosis models play a pivotal role in enabling healthcare practitioners to administer timely and accurate interventions for sepsis, thereby augmenting patient outcomes. Nevertheless, the majority of available models consider the overall physiological attributes of patients, overlooking the asynchronous spatiotemporal interactions among multiple organ systems. These constraints hinder a full application of such models, particularly when dealing with limited clinical data. To surmount these challenges, a comprehensive model, denoted as recurrent Graph Attention Network-multi Gated Recurrent Unit (rGAT-mGRU), was proposed. Taking into account the intricate spatiotemporal interactions among multiple organ systems, the model predicted in-hospital mortality of sepsis using data collected within the 48-hour period post-diagnosis. Material and methodsMultiple parallel GRU sub-models were formulated to investigate the temporal physiological variations of single organ systems. Meanwhile, a GAT structure featuring a memory unit was constructed to capture spatiotemporal connections among multi-organ systems. Additionally, an attention-injection mechanism was employed to govern the data flowing within the network pertaining to multi-organ systems. The proposed model underwent training and testing using a dataset of 10,181 sepsis cases extracted from the Medical Information Mart for Intensive Care III (MIMIC-III) database. To evaluate the model's superiority, it was compared with the existing common baseline models. Furthermore, ablation experiments were designed to elucidate the rationale and robustness of the proposed model. ResultsCompared with the baseline models for predicting mortality of sepsis, the rGAT-mGRU model demonstrated the largest area under the receiver operating characteristic curve (AUROC) of 0.8777 ± 0.0039 and the maximum area under the precision-recall curve (AUPRC) of 0.5818 ± 0.0071, with sensitivity of 0.8358 ± 0.0302 and specificity of 0.7727 ± 0.0229, respectively. The proposed model was capable of delineating the varying contribution of the involved organ systems at distinct moments, as specifically illustrated by the attention weights. Furthermore, it exhibited consistent performance even in the face of limited clinical data. ConclusionThe rGAT-mGRU model has the potential to indicate sepsis prognosis by extracting the dynamic spatiotemporal interplay information inherent in multi-organ systems during critical diseases, thereby providing clinicians with auxiliary decision-making support.

Pose-Aided Video-Based Person Re-Identification via Recurrent Graph Convolutional Network

Existing methods for video-based person re-identification (ReID) mainly learn the appearance feature of a given pedestrian via a feature extractor and a feature aggregator. However, the appearance models would fail to learn a large inter-class variance when different pedestrians have similar appearances. Considering that different pedestrians have different walking postures and body proportions, we propose to learn the discriminative pose feature beyond the appearance feature for video retrieval. Specifically, we implement a two-branch architecture to separately learn the appearance feature and pose feature, and then concatenate them together for inference. To learn the pose feature, we first detect the pedestrian pose in each frame through an off-the-shelf pose detector, and construct a temporal graph using the pose sequence. We then exploit a recurrent graph convolutional network (RGCN) to learn the node embeddings of the temporal pose graph, which devises a global information propagation mechanism to simultaneously achieve the neighborhood aggregation of intra-frame nodes and message passing among inter-frame graphs. Finally, we propose a dual-attention method (DAM) consisting of node-attention and time-attention to obtain the temporal graph representation from the node embeddings, where the self-attention mechanism is employed to learn the importance of each node and each frame. We verify the proposed method on three video-based ReID datasets, i.e., Mars, DukeMTMC and iLIDS-VID, whose experimental results demonstrate that the learned pose feature can effectively improve the performance of existing appearance models.

Deep Recurrent Graph Convolutional Architecture for Sensor Fault Detection, Isolation, and Accommodation in Digital Twins

Deep Recurrent Graph Convolutional Architecture for Sensor Fault Detection, Isolation, and Accommodation in Digital Twins

Micro video recommendation in multimodality using dual-perception and gated recurrent graph neural network

Micro video recommendation in multimodality using dual-perception and gated recurrent graph neural network

Metro Station functional clustering and dual-view recurrent graph convolutional network for metro passenger flow prediction

The metro system is indispensable for alleviating traffic congestion in the urban transportation system. Precise metro passenger flow (MPF) prediction is crucial in ensuring smooth operations of the metro system. Recently, the graph convolutional network (GCN), which is effective in the spatial feature extraction, has been applied in traffic prediction. However, most existing GCN-based methods construct the empirical graphs based on distance and adjacency, which cannot fully express the correlations of metro stations. This paper proposes a novel MPF prediction method consisting of three parts: K-means-based metro station functional clustering (KMSFC), external feature fusion, and dual-view recurrent GCN (DVRGCN). The KMSFC identifies the metro stations both having similar MPF changing tendencies and being located in similar urban functional areas. Furthermore, the DVRGCN is designed to simultaneously capture the spatiotemporal and external features. The dual-view GCN module in the DVRGCN captures both explicit and implicit spatial features of the metro traffic network. To demonstrate the capability for making accurate MPF predictions, the experiments using a real-world metro traffic dataset are conducted. The ablation experiments are also performed to prove the contribution of each module in the proposed method. The experimental results show that the proposed method outperforms other state-of-the-art traffic prediction methods.

A Deep Learning Approach to the Prediction of Drug Side-Effects on Molecular Graphs.

Predicting drug side effects before they occur is a critical task for keeping the number of drug-related hospitalizations low and for improving drug discovery processes. Automatic predictors of side-effects generally are not able to process the structure of the drug, resulting in a loss of information. Graph neural networks have seen great success in recent years, thanks to their ability of exploiting the information conveyed by the graph structure and labels. These models have been used in a wide variety of biological applications, among which the prediction of drug side-effects on a large knowledge graph. Exploiting the molecular graph encoding the structure of the drug represents a novel approach, in which the problem is formulated as a multi-class multi-label graph-focused classification. We developed a methodology to carry out this task, using recurrent Graph Neural Networks, and building a dataset from freely accessible and well established data sources. The results show that our method has an improved classification capability, under many parameters and metrics, with respect to previously available predictors. The method is not ready for clinical tests yet, as the specificity is still below the preliminary 25% threshold. Future efforts will aim at improving this aspect.

The Large (<i>M</i> ≥ 6.0) Earthquakes Seismic Precursors in the Junction Zone of the Kuril-Kamchatka and Aleutian Island Arcs

The article describes the behavior of the series of seismic precursors before the activation of 2016‒2019 in an area with a tectonically complex geodynamic environment at the boundaries of the Pacific, North American, and Eurasian lithospheric plates. In order to clarify the location of possible focal areas of future strong earthquakes, a comprehensive approach to the application of allocated precursors was considered: quiescence by parameters RTL and Z-function, variations of the slope of the recurrence graph γ and areas of seismic ruptures dS. For the first time, in relation to Kamchatka, for anomalies to parameter dS, interpreted as seismic activations, the presence of a predictive component before the onset of foreshock activation is shown by the example of an increase in its values before the events that are the strongest for the areas where their focal areas are located: the Near islands Aleutian earthquake with Mw = 7.8 on July 17, 2017 and the Uglovoye Podnyatiye earthquake on December 20, 2018, Мw = 7.3, is demonstrated. The total length of the anomalous region, currently existing and including successive stages of quiescences and foreshock activations is ~900 km, which allows for the possibility of a new strong earthquake with a magnitude exceeding all previously recorded in this area in the junction zone of three lithospheric plates.

A3GC-IP: Attention-oriented adjacency adaptive recurrent graph convolutions for human pose estimation from sparse inertial measurements

Conventional methods for human pose estimation either require a high degree of instrumentation, by relying on many inertial measurement units (IMUs), or constraint the recording space, by relying on extrinsic cameras. These deficits are tackled through the approach of human pose estimation from sparse IMU data. We define attention-oriented adjacency adaptive graph convolutional long-short term memory networks (A3GC-LSTM), to tackle human pose estimation based on six IMUs, through incorporating the human body graph structure directly into the network. The A3GC-LSTM combines both spatial and temporal dependency in a single network operation, more memory efficiently than previous approaches. The recurrent graph learning on arbitrarily long sequences is made possible by equipping graph convolutions with adjacency adaptivity, which eliminates the problem of information loss in deep or recurrent graph networks, while it also allows for learning unknown dependencies between the human body joints. To further boost accuracy, a spatial attention formalism is incorporated into the recurrent LSTM cell. With our presented approach, we are able to utilize the inherent graph nature of the human body, and thus can outperform the state-of-the-art for human pose estimation from sparse IMU data.

Query Path Generation via Bidirectional Reasoning for Multihop Question Answering From Knowledge Bases

Multi-hop question answering from knowledge bases (KBQA) is a hot research topic in natural language processing. Recently, graph neural network-based (GNN-based) methods have achieved promising results as the KB can be organized as a knowledge graph (KG). However, they often suffered from the sparsity of the KG which was detrimental to the structure-encoding and reasoning capabilities of GNN. Specifically, a KG is a sparse graph linked by directed relations and previous studies have paid scant attention to the directional characteristic of relations in the KG, limiting the patterns of relation path that GNN-based approaches could resolve. This study proposes a bidirectional recurrent graph neural network (BRGNN) to tackle these difficulties. To model the bidirectional information of relations, all adjacent relations of an entity are grouped by their directions, and they are separately aggregated into the entity representation in outward and inward directions. For the reasoning process, BRGNN simultaneously considers the neighbor relations in both directions to cover more patterns of relation paths and improve the recall of answers. Extensive experiments on three benchmarks: WebQuestionsSP, ComplexWebQuestions and MetaQA, verify that BRGNN can answer more questions by taking into account the directional information, and it is competitive to all state-of-the-art approaches.

A Learnable Image-Based Load Signature Construction Approach in NILM for Appliances Identification

One of the tasks of Non-Intrusive Load Monitoring (NILM) is load identification, which aims to extract and classify altered electrical signals after switching events are detected. In this subtask, representative and distinguishable load signatures are es-sential. At present, the literature approach to characterize electri-cal appliances is mainly based on manual feature engineering. However, the performance of signatures obtained by this way is limited. In this paper, we propose a novel load signature construc-tion method utilizing deep learning techniques. Specifically, three learnable load signatures are presented such as Learnable Recur-rent Graph (LRG), Learnable Gramian Matrix (LGM) and Gen-erative Graph (GG). Furthermore, we test different frameworks for learning these signatures and conclude that Temporal Convo-lutional Networks (TCN) based on residual learning are more suit-able for this work than the other schemes mentioned. The results of experiment on the PLAID datasets with submetered and aggre-gated, WHITED dataset and LILAC dataset confirm that our method outperforms the voltage-current trajectory, Recursive Graph and Gramian Angular Field methods in multiple evalua-tion metrics.

From signals to music: a bottom-up approach to the structure of neuronal activity.

The search for the "neural code" has been a fundamental quest in neuroscience, concerned with the way neurons and neuronal systems process and transmit information. However, the term "code" has been mostly used as a metaphor, seldom acknowledging the formal definitions introduced by information theory, and the contributions of linguistics and semiotics not at all. The heuristic potential of the latter was suggested by structuralism, which turned the methods and findings of linguistics to other fields of knowledge. For the study of complex communication systems, such as human language and music, the necessity of an approach that considers multilayered, nested, structured organization of symbols becomes evident. We work under the hypothesis that the neural code might be as complex as these human-made codes. To test this, we propose a bottom-up approach, constructing a symbolic logic in order to translate neuronal signals into music scores. We recorded single cells' activity from the rat's globus pallidus pars interna under conditions of full alertness, blindfoldedness and environmental silence. We analyzed the signals with statistical, spectral, and complex methods, including Fast Fourier Transform, Hurst exponent and recurrence plot analysis. The results indicated complex behavior and recurrence graphs consistent with fractality, and a Hurst exponent >0.5, evidencing temporal persistence. On the whole, these features point toward a complex behavior of the time series analyzed, also present in classical music, which upholds the hypothesis of structural similarities between music and neuronal activity. Furthermore, through our experiment we performed a comparison between music and raw neuronal activity. Our results point to the same conclusion, showing the structures of music and neuronal activity to be homologous. The scores were not only spontaneously tonal, but they exhibited structure and features normally present in human-made musical creations. The hypothesis of a structural homology between the neural code and the code of music holds, suggesting that some of the insights introduced by linguistic and semiotic theory might be a useful methodological resource to go beyond the limits set by metaphoric notions of "code."

Fine-Gained Recurrence Graph: Graphical Modeling of Vibration Signal for Fault Diagnosis of Wind Turbine

Benefiting from the recent successes of convolutional neural networks (CNNs), many studies have modeled the vibration signal of energy system into a two-dimensional (2D) input graph to amplify and highlight fault features. However, most works seldomly consider the system dynamic characteristics, which may affect the knowledge discovery and diagnosis quality. To address this issue, this paper proposes a novel graphical modeling approach, termed as fine-gained recurrence graph (FRG), to capture dynamic characteristics of vibration signals from the view of nonlinear dynamics. FRG focuses on modeling the temporal correlations and tendencies between state vectors in phase space and then visualizes the characteristics to represent intrinsic dynamic changes of the vibration signals into 2D graphs. The information representation capability of the proposed FRG is verified using different dynamic signals. Based on this finding, an ensemble fault diagnosis model is proposed fusing FRG with deep CNNs. Meanwhile, transfer learning is accompanied to deal with the training difficulties of deep CNNs. Finally, case studies on experimental data and real wind turbine data illustrate its effectiveness and feasibility. Comparisons with four state-of-the-art approaches have confirmed the preferable information representation and diagnosis performance of the proposed approach.

Multiple-model and time-sensitive dynamic active learning for recurrent graph convolutional network model extraction attacks

Multiple-model and time-sensitive dynamic active learning for recurrent graph convolutional network model extraction attacks

A knowledge-based task planning approach for robot multi-task manipulation

Task planning is a crucial component in facilitating robot multi-task manipulations. Language-based task planning methods offer practicality in receiving commands from humans in real-life scenarios and require only low-cost labeled data. However, existing methods often rely on sequence models for planning, which primarily focus on mapping language to sequences of sub-tasks while neglecting the knowledge about tasks and objects. To overcome these limitations, we propose a knowledge-based task planning approach called Recurrent Graph Convolutional Network (RGCN). It is devised with a novel structure that combined GCN (Kipf and Welling in International Conference on Learning Representations (ICLR), 2017) and LSTM (Hochreiter and chmidhuber in Neural Comput 9 (8): 1735-1780, 1997. https://doi.org/10.1162/neco.1997.9.8.1735) which enables it to leverage knowledge graph data and historical predictions. The experimental results demonstrate that our approach achieves the impressive task planning success rate of 95.7%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$${95.7\\%}$$\\end{document}, surpassing the best baseline method significantly, which achieves 78.7%\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$${78.7\\%}$$\\end{document}. Furthermore, we evaluate the performance of multi-task manipulation across a specific set of 20 tasks within a simulated environment. Notably, RGCN combined with pre-trained primitive tasks exhibits the highest success rate compared with state-of-art multi-task learning methods. Our method is proven to be significant for language-conditioned task planning and is qualified for instructing robots for multi-task manipulation.

RGCNU: Recurrent Graph Convolutional Network With Uncertainty Estimation for Remaining Useful Life Prediction

Dear Editor, This letter focuses on the problem of remaining useful life (RUL) prediction of equipment. Existing graph neural network (GCN)-based approaches merely provide the point estimation of RUL. However, the estimated RUL often varies widely due to the model parameters and the noise in data. It is important to know the uncertainty in predictions for reliable risk analysis and maintenance decision making. To map the relationship between noisy condition monitoring data and RUL with uncertainty, we propose a recurrent graph convolutional network with uncertainty estimation (RGCNU) for RUL prediction. In our approach, the correlation exploiting module captures the spatial-temporal correlations based on the learned graph structure. Furthermore, the fusion module associates the RUL prediction and data uncertainty to improve the robustness of the model to noisy data.

Multi-UAV Navigation for Partially Observable Communication Coverage by Graph Reinforcement Learning

In this paper, we aim to design a deep reinforcement learning (DRL) based control solution to navigating a swarm of unmanned aerial vehicles (UAVs) to fly around an unexplored target area under partial observation, which serves as Mobile Base Stations (MBSs) providing optimal communication coverage for the ground mobile users. To handle the information loss caused by the partial observability, we introduce a novel network architecture named Deep Recurrent Graph Network (DRGN), which could obtain extra spatial information through graph-convolution based inter-UAV communication, and utilize historical features with a recurrent unit. Based on DRGN and maximum-entropy learning, we propose a stochastic DRL policy named Soft Deep Recurrent Graph Network (SDRGN). In SDRGN, a heuristic reward function is elaborated, which is based on the local information of each UAV instead of the global information; thus, SDRGN reduces the training cost and enables distributed online learning. We conducted extensive experiments to design the structure of DRGN and examine the performance of SDRGN. The simulation results show that the proposed model outperforms four state-of-the-art DRL-based approaches and three heuristic baselines, and demonstrate the scalability, transferability, robustness, and interpretability of SDRGN.