Network-based Method Research Articles

A longitudinal study of the brain structure network changes in HIV patients with ANI: combined VBM with SCN.

Despite the widespread adoption of combination antiretroviral therapy (cART) in managing HIV, the virus's impact on the brain structure of patients remains significant. This study aims to longitudinally explore the persistent effects of HIV on brain structure, focusing on changes in gray matter volume (GMV) and structural covariance network (SCN) among patients at the Asymptomatic Neurocognitive Impairment (ANI) stage. This research involved 45 HIV patients diagnosed with ANI and 45 demographically matched healthy controls (HCs). The participants were observed over a 1.5-year period. Differences in GMV between groups were analyzed using voxel-based morphometry (VBM), while the graph theory model facilitated the establishment of topological metrics for assessing network indices. These differences were evaluated using two-sample t-tests and paired-sample t-tests, applying the network-based statistics method. Additionally, the study examined correlations between GMV and cognitive performance, as well as clinical variables. Compared with HCs, HIV patients demonstrated reduced GMV in the right middle temporal gyrus and left middle frontal gyrus (FWE, p < 0.05), along with decreased betweenness centrality (BC) in the left anterior cingulate and paracingulate cortex. Conversely, an increase in the clustering coefficient (Cp) was observed (FDR, p < 0.05). During the follow-up period, a decline in GMV in the right fusiform gyrus (FWE, p < 0.05) and a reduction in node efficiency (Ne) in the triangular part of the inferior frontal gyrus were noted compared with baseline measurements (FDR, p < 0.05). The SCN of HIV patients exhibited small-world properties across most sparsity levels (Sigma >1), and area under the curve (AUC) analysis revealed no significant statistical differences between groups. The findings suggest that despite the administration of combination antiretroviral therapy (cART), HIV continues to exert slow and sustained damage on brain structures. However, when compared to HCs, the small-world properties of the patients' SCNs did not significantly differ, and the clustering coefficient, indicative of the overall information-processing capacity of the brain network, was slightly elevated in HIV patients. This elevation may relate to compensatory effects of brain area functions, the impact of cART, functional reorganization, or inflammatory responses.

A Data-Driven Machine Learning Approach for Turbulent Flow Field Prediction Based on Direct Computational Fluid Dynamics Database

A novel approach is presented for predicting compressible turbulent flow fields using a neural network-based data-driven method. Accurate prediction in turbulent regions heavily relies on the resolution of available data. Traditional methods, employing image-based techniques by mapping scattered computational fluid dynamics (CFD) data onto Cartesian grids, encounter data scarcity in critical areas such as the boundary layer and wake. Recently, convolutional neural networks (CNN) have gained prominence as the most widely referenced technique in fluid dynamics, utilizing flow field images as datasets for flow field prediction. However, CNN requires datasets with a high pixel density to enhance training accuracy in crucial regions, thereby increasing the input data volume and machine training time. To address this challenge, our proposed method deviates from using flow field images and instead generates datasets directly from the flow field properties of CFD grid points. By employing this approach, several advantages are realized. Firstly, the network benefits from the favorable characteristics of unstructured grids, such as varying point spacing near the object surface and in the far field, which effectively reduces the amount of input data and consequently the machine training cost. Secondly, the construction of the training dataset eliminates the need for interpolation or extrapolation, thereby preserving the accuracy of CFD data. In this case, a simple multilayer perceptron can be trained using the proposed dataset. Various flow field properties, including static pressure, turbulent kinetic energy, and velocity components, can be predicted with high accuracy within a few seconds.

Bayesian Neural Network-Based Equipment Operational Trend Prediction Method Using Channel Attention Mechanism

Bayesian Neural Network-Based Equipment Operational Trend Prediction Method Using Channel Attention Mechanism

A Novel Neural Network-Based Control Method for Proton Exchange Membrane Fuel Cell in DC Microgrids

A Novel Neural Network-Based Control Method for Proton Exchange Membrane Fuel Cell in DC Microgrids

Heterogeneous Network-Based Contrastive Learning Method for PolSAR Land Cover Classification

Heterogeneous Network-Based Contrastive Learning Method for PolSAR Land Cover Classification

Lung image segmentation via generative adversarial networks.

Lung image segmentation plays an important role in computer-aid pulmonary disease diagnosis and treatment. This paper explores the lung CT image segmentation method by generative adversarial networks. We employ a variety of generative adversarial networks and used their capability of image translation to perform image segmentation. The generative adversarial network is employed to translate the original lung image into the segmented image. The generative adversarial networks-based segmentation method is tested on real lung image data set. Experimental results show that the proposed method outperforms the state-of-the-art method. The generative adversarial networks-based method is effective for lung image segmentation.

Uncertainty quantification in neural-network based pain intensity estimation.

Improper pain management leads to severe physical or mental consequences, including suffering, a negative impact on quality of life, and an increased risk of opioid dependency. Assessing the presence and severity of pain is imperative to prevent such outcomes and determine the appropriate intervention. However, the evaluation of pain intensity is a challenging task because different individuals experience pain differently. To overcome this, many researchers in the field have employed machine learning models to evaluate pain intensity objectively using physiological signals. However, these efforts have primarily focused on pain point estimation, disregarding inherent uncertainty and variability in the data and model. A point estimate, which provides only partial information, is not sufficient for sound clinical decision-making. This study proposes a neural network-based method for objective pain interval estimation, and quantification of uncertainty. Our approach, which enables objective pain intensity estimation with desired confidence probabilities, affords clinicians a better understanding of a person's pain intensity. We explored three distinct algorithms: the bootstrap method, lower and upper bound estimation (LossL) optimized by genetic algorithm, and modified lower and upper bound estimation (LossS) optimized by gradient descent algorithm. Our empirical results demonstrate that LossS outperforms the other two by providing narrower prediction intervals. For 50%, 75%, 85%, and 95% prediction interval coverage probability, LossS provides average interval widths that are 22.4%, 7.9%, 16.7%, and 9.1% narrower than those of LossL, and 19.3%, 21.1%, 23.6%, and 26.9% narrower than those of bootstrap. As LossS outperforms, we assessed its performance in three different model-building approaches: (1) a generalized approach using a single model for the entire population, (2) a personalized approach with separate models for each individual, and (3) a hybrid approach with models for clusters of individuals. Results demonstrate that the hybrid model-building approach provides the best performance.

Predicting Braess's paradox of power grids using graph neural networks.

As an increasing number of renewable energy generators are integrated into the electrical grid, the necessity to add new transmission lines to facilitate power transfer and ensure grid stability becomes paramount. However, the addition of new transmission lines to the existing grid topology can lead to the emergence of Braess's paradox or even trigger grid failures. Hence, predicting where to add transmission lines to guarantee stable grid operation is of utmost importance. In this context, we employ deep learning to address this challenge and propose a graph neural network-based method for predicting Braess's paradox in electrical grids, framing the problem of adding new transmission lines causing Braess's paradox as a graph classification task. Taking into consideration the topological and electrical attributes of the grid, we select node features such as degree, closeness centrality, and power values. This approach assists the model in better understanding the relationships between nodes, enhancing the model's representational capabilities. Furthermore, we apply layered adaptive weighting to the output of the graph isomorphism network to emphasize the significance of hierarchical information that has a greater impact on the output, thus improving the model's generalization across electrical grids of varying scales. Experimental results on the IEEE 39, IEEE 57, and IEEE 118 standard test systems demonstrate the efficiency of the proposed method, achieving prediction accuracies of 93.8%, 88.8%, and 88.1%, respectively. Model visualization and ablation studies further validate the effectiveness of this approach.

Computer vision enabled high-quality electrochemical experimentation

We have developed a convolutional neural network-based method that utilizes microscopic images of prepared electrodes to directly assess sample quality. This method paves the way towards high-quality, automated electrochemical experimentation.

A Hierarchical Network-Based Method for Predicting Driver Traffic Violations

A Hierarchical Network-Based Method for Predicting Driver Traffic Violations

A Multi-scale Hierarchical Residual Network-based Method for Tiny Object Detection in Optical Remote Sensing Images

A Multi-scale Hierarchical Residual Network-based Method for Tiny Object Detection in Optical Remote Sensing Images

Revelando desigualdades espaciais de renda e raciais e padrões de segregação com um indicador configuracional de acessibilidade potencial.

Abstract Urban segregation has been a central challenge in the pursuit of equity and sustainability, especially in Latin American cities. The existing body of research on segregation, however, has predominantly focused on cities in the United States and Europe. Most existing metrics on segregation do not usually explicitly address the spatial dimension and diversity of cities’ spatial structures. As a result, there is still a gap in the methods to properly capture the spatial relationships implied in segregation. This paper proposes and applies a network-based method for assessing the inequalities in access to opportunities for different social groups (income and race), and the patterns of socio-spatial segregation based on urban accessibility levels. We apply the method to the city of Pelotas, Brazil, combining the street network configuration with population densities and spatial distribution of urban opportunities. Results revealed that high-income white groups dominate the accessibility levels, benefiting from a concentration of opportunities in the central district, while low-income non-white groups experience poorer accessibility. The method allowed to uncover a center-periphery segregation pattern, although more complex and spatially diversified, providing valuable insights that can inform urban planning and policy-making efforts at the intraurban scale.

Parameter Precise Estimation Technology of Active Segment of Non-cooperative Targets Based on Long Short-Term Memory

Traditional algorithms do not fully utilize the timing information of non-cooperative targets, and setting too many motion parameters can lead to complex dynamic model calculations. This paper proposes a long short-term memory (LSTM) network-based method for estimating the parameters of the active segment of the non-cooperative target under single-satellite observation. Based on the simulation training set of the active segment of the non-cooperative target, the network parameters of the LSTM network are designed, the motion characteristics of the active segment of the non-cooperative target are fully excavated through data-driven methods, and the candidate cutting trajectories are screened and predicted to realize the estimation of the motion parameters of the active segment of the non-cooperative target under the condition of single-satellite observation. The experimental results show that the estimation method proposed in this paper can effectively deal with the inaccurate problem with the non-cooperative target’s active segment motion model established under the condition of single-satellite observation, obtain more accurate active segment motion parameters, and provide a feasible new idea and method for the parameter estimation of the active segment of the non-cooperative target under the single-satellite observation.

Rumor Localization, Detection and Prediction in Social Network

With the global epidemic of the COVID-19, various rumors spread wantonly on social networks, which has seriously affected the stability and harmony of the entire society. To purify the network environment, some researchers have proposed to fight rumors from the perspectives of tracing the source of rumors, detecting the authenticity of information, and predicting explosive fake news. But their works are fragmented, and their performance are not significant. So we need strong antirumor methods to fight rumors. To this end, this article proposes a more comprehensive antirumor mechanism, which can realize rumors source location, rumor detection, and popularity prediction (RLDP). In particular, in the task of localization, we propose graph neural network-based method, which does not need to specify the underlying propagation mode and the number of rumor sources; in the task of detection, utilizing lightGBM, we construct a rumor detection model; in the task of popularity prediction, we construct a model based on contrastive learning while considering user engagements and information propagation, and the text of rumor. Finally, we verify the performance of the proposed RLDP by conducting extensive experiments.

Human Behavior Recognition Based on CNN-LSTM Hybrid and Multi-Sensing Feature Information Fusion

To address the human activity recognition problem and its application in practical situations, a CNN-LSTM hybrid neural network model capable of automatically extracting sensor data features and memorizing temporal activity data is designed and improved by integrating CNN and gated recurrent units as a variant of RNN. A multi-channel spatiotemporal fusion network-based two-person interaction behavior recognition method is proposed for two-person skeletal sequential behavior recognition. Firstly, a viewpoint invariant feature extraction method is used to extract two-player skeleton features, then a two-layer cascaded spatiotemporal fusion network model is designed, and finally, a multi-channel spatiotemporal fusion network is used to learn multiple sets of two-player skeleton features separately to obtain multi-channel fusion features, and the fusion features are used to recognize the interaction behavior, and the weights are shared among the channels. Applying the algorithm in the paper to the UCF101 dataset for experiments, the accuracy of the two-person cross-object experiment can reach 96.42% and the accuracy of the cross-view experiment can reach 97.46%. The method in the paper shows better performance in two-player interaction behavior recognition compared to typical methods in this field.

Parameter inversion method for target plate constitutive model based on BP neural network

In this paper, a neural network-based inversion method that can be used to streamline the acquisition of polymethyl methacrylate (PMMA) material constitutive model parameters for explosive cutting simulations is presented. This approach mitigates the need for multiple tests, a requirement of traditional methods. Initially, three categories of the damage, penetration depth, impact fracture thickness and spallation damage thickness, were distinguished and quantified by cutting the14 mm PMMA target plate with a 2.5 mm wide linear shaped charge, and the adjustment interval for the constitutive model parameters could be determined by combining the explosive cutting experiment data and the empirical parameters of the Johnson Holmquist Ceramics (JH-2) constitutive model. Subsequently, the numerical simulations of the cutting process of the performed target plate tests were carried out using LS-DYNA, and a dataset of target plate damage incorporating the three types of damage data was amassed. Ultimately, a neural network model correlating the parameters of the PMMA target plate constitutive model with the damage data was developed and trained using the plate damage dataset, and several supplementary experiments as well as finite element numerical simulations involving a 4.2 mm wide linear shaped charge cutting a 19 mm PMMA target plate were conducted. The numerical results displayed minimal variance from the experimental results in terms of fracture characteristics and damage data, suggesting that the inverted JH-2 constitutive model parameters can be effectively applied to PMMA target plate explosive cutting simulations. The parameter inversion method allows for the acquisition of more accurate material constitutive model parameters with fewer experiments and will be beneficial for further theoretical studies and practical applications of target plates in the future.

Exploring the impact of global competitiveness pillars on sustainable development

Global competitiveness refers to a nation's ability to boost its economic growth and advancement. The Global Competitiveness Index (GCI), introduced by the World Economic Forum, serves as a tool to evaluate how individual countries perform across 12 key competitiveness pillars. United Nations member states are committed to achieving the Sustainable Development Goals (SDGs) by 2030, addressing economic, social, and environmental aspects of sustainability. This research employs a data-driven Bayesian Belief Network-based method to identify crucial GCI pillars, considering their positive and negative impacts on sustainability performance. Notably, ‘health’, ‘infrastructure’, and ‘skills’ are identified as pivotal GCI pillars with the potential to adversely affect sustainability performance. Conversely, ‘innovation capability’, ‘labor market’, and ‘skills’ are highlighted for their positive influence on overall sustainability performance. This is the first study to explore the link between competitiveness pillars and sustainability performance in a probabilistic network framework, providing valuable insights for policymakers aiming to enhance both competitiveness and sustainability.

Genetic Discrimination of Grade 3 and Grade 4 Gliomas by Artificial Neural Network.

Gliomas, including anaplastic gliomas (AG; grade 3) and glioblastomas (GBM; grade 4), are malignant brain tumors associated with poor prognosis and low survival rates. Current classification systems based on histopathology have limitations due to intratumoral heterogeneity. The treatment and prognosis are distinctly different between grade 3 and grade 4 gliomas patients. Therefore, there is a need for molecular markers to differentiate these tumors accurately. In this study, we aimed to identify a gene expression signature using an artificial neural network (ANN) in application to microarray and serial analysis of gene expression (SAGE) data for grade 3 (AG) and grade 4 (GBM) gliomas discrimination. We acquired gene expression data from publicly available datasets on glial tumors of grades 3 and 4-a total of 93 grade 3 gliomas and 224 grade 4 gliomas. To select genes for classification, we implemented an artificial neural network-based method using a combination of self-organized maps (SOM) and perceptron. In general, we implemented a multi-stage procedure that involved multiple runs of a genetic algorithm to identify genes that provided optimal clusterization on the SOM. We performed this procedure multiple times, resulting in different sets of genes each time. Eventually, we selected several genes that appeared most frequently in the reduced sets and performed classification using them. Our analysis identified a set of seven genes (BCAS4, GLUD2, KCNJ10, KCND2, AKR7A2, FOLR1, and KIAA0319). The classification accuracy using this gene set was 87.5%. These findings suggest the potential of this gene set as a molecular marker for distinguishing grade 3 (AG) from grade 4 (GBM) gliomas.

Investigation of CMP Process of Silica Substrate by Ceria Abrasive: Insights from Molecular Dynamics Simulation with Neural Network Potential

Chemical mechanical polishing (CMP) is a process of surface planarization by chemical and mechanical driving forces. CMP is expected to solve various problems, including the surface roughness on semiconductor devices. An aqueous slurry, including abrasive grains and chemical components, is used in the CMP process. Development of slurries with high polishing rate, excellent planarity, and superior processability is not only a key challenge, but also a significant opportunity for the future of our industry. In order to achieve the development, it is necessary to understand the detailed mechanism of CMP, which involves complex phenomena accompanied by interfacial chemical reactions and mechanical friction. Computational scientific approaches, such as density functional theory (DFT) and molecular dynamics (MD) simulations, are powerful tools for understanding complex tribochemical phenomena at the atomic scale. However, these methods require much computation time and present difficulties for large-scale simulations. To overcome this limitation, a computational method called Neural Network Potential (NNP) is a powerful tool because it achieves a good balance between accuracy and computational efficiency by constructing potential energy surfaces based on a neural network-based method. We performed NNP-MD simulations of the ceria/silica interface with the development of CMP slurry including ceria abrasive grains, which is typically used on the surfaces of silicon layered devices. As a result, we succeeded in observing polishing phenomena at the interface at the atomic scale, namely Ce-O-Si bond formation between ceria abrasive grains and silica substrate, hydrolysis reaction of Si-O bond, and associated mono-silica dissociation from the substrate. We also performed simulations of systems containing chemical compounds added to the slurry for the purpose of controlling the polishing rate and observed the role of additives on the CMP process. In this presentation, we will show simulation results and discuss the mechanism of CMP of silica substrates with ceria abrasive grains in terms of dependence on surface morphology at the atomic scale and the role of additives. Reference [1] Tran*, R., Lan*, J., Shuaibi*, M., Wood*, B., Goyal*, S., Das, A., Heras-Domingo, J., Kolluru, A., Rizvi, A., Shoghi, N., Sriram, A., Ulissi, Z., & Zitnick, C. (2022). The Open Catalyst 2022 (OC22) Dataset and Challenges for Oxide Electrocatalysis. arXiv preprint arXiv:2206.08917.

Rapid three-dimensional detection of harmful algae using holographic microimaging

Holographic microimaging technology is widely used in the plankton detection due to its large depth of field and ability to record three-dimensional (3D) information. However, traditional methods involve tedious and time-consuming steps, e.g., region of interest (ROI) extraction, dense reconstruction, focused evaluation, and focused ROI recognition. As a result, traditional methods are difficult to detect in real time. In this paper, we propose a neural network-based 3D detection method, which for the first time uses the holograms to obtain 3D sampling data for algae detection, and innovatively designs a convolutional neural network (DHCNN) to improve the speed and accuracy of detection. A dataset containing 3D samples of eight harmful algae species is established, and the trained DHCNN model achieves a mean average precision of 99.2 % on the test set. We find that accuracy is higher with 3D sampled data, which contains more valid information than with autofocused ROIs. Moreover, detection rate is 99.5 %, which means it could detect attached algae and algae abundance statistics are more accurate. And detection time of the proposed method is only 0.066 s, which is 20 times faster than the traditional method. We believe that the proposed method represents an effective approach for real-time algae monitoring.