Flow Datasets Research Articles

Global Context Volume construction and semantics-guided disparity refinement for stereo matching

The accuracy of stereo matching has been greatly improved with the advent of convolutional neural networks. However, existing methods still perform poorly in regions of low-texture and reflections due to insufficient matching information. In this paper, we propose a novel semantic stereo matching network named GSSNet, which combines global context information with high-level semantic clues to further enrich the matching features. Specifically, we employ Swin Transformer as a joint feature extractor to capture multi-scale global features for both semantic segmentation and disparity estimation. By fusing low-resolution multi-scale feature maps progressively, we construct a Global Context Volume (GCV) representation covering long-range receptive field for initial disparity estimation. Initial disparities are further refined with a high-level embedding learned from the semantic segmentation branch of the network through residual learning. During training, due to the lack of joint annotations, we also propose a simple and effective pseudo-labeling strategy. Comprehensive experimental results demonstrated on various datasets manifest the effectiveness of the proposed GSSNet. Among all published methods as of 9, February 2024, our approach ranks 1st on KITTI 2012 leaderboard and 3rd on KITTI 2015 leaderboard, and produces competitive results on Scene Flow and Cityscapes datasets. Code will be available at https://github.com/Twil-7/GSSNet.

Neural fields for rapid aircraft aerodynamics simulations

This paper presents a methodology to learn surrogate models of steady state fluid dynamics simulations on meshed domains, based on Implicit Neural Representations (INRs). The proposed models can be applied directly to unstructured domains for different flow conditions, handle non-parametric 3D geometric variations, and generalize to unseen shapes at test time. The coordinate-based formulation naturally leads to robustness with respect to discretization, allowing an excellent trade-off between computational cost (memory footprint and training time) and accuracy. The method is demonstrated on two industrially relevant applications: a RANS dataset of the two-dimensional compressible flow over a transonic airfoil and a dataset of the surface pressure distribution over 3D wings, including shape, inflow condition, and control surface deflection variations. On the considered test cases, our approach achieves a more than three times lower test error and significantly improves generalization error on unseen geometries compared to state-of-the-art Graph Neural Network architectures. Remarkably, the method can drastically accelerate high fidelity numerical solver, achieving a five order of magnitude speedup on the RANS transonic airfoil dataset.

Financial, institutional, and macroeconomic determinants of cross-country portfolio equity flows: The case of developed countries

This paper examines the determinants of financial equity flows to investigate the role played by business cycles, government debt and sovereign rating scores, and whether the impact depends on the magnitude and direction of the flows. Using a new, richer dataset of flows among developed countries over 2001–2018, our key findings are as follows: (i) equity flows are more intense among countries at the same stage of the business cycle (ii) equity flows are higher to countries with a relatively lower debt to GDP ratio (iii) financial and macroeconomic variables are important for big equity flows, while institutional variables are important for the small flows. Overall, considering a wider range of factors under-explored in the literature, we provide a stronger understanding of the development of risks in the financial sector as well as the linkages with other sectors of the economy.

Stereo disparity estimation based on Siamese network

Abstract This paper proposes an approach for estimating stereo disparity based on a Siamese network. The existing Top-k disparity regression strategy may overlook the true disparity at object edges. To address this issue, an additional maximum disparity estimation value is introduced to optimize the Top-k strategy, creating the Top-k+ strategy. Furthermore, this research improves the MobileNetv2 (MV2) block by introducing an attention mechanism in the frequency domain, resulting in a more efficient FMV2 block that extracts high-frequency information such as textures and edges. Considering the excellent performance of the gated recurrent unit (GRU) in disparity optimization, the algorithm adopts an iterative optimization method based on GRU. Comparative studies pertaining to the Scene Flow, KITTI 2015, and ETH3D benchmark datasets show outstanding results in disparity estimation, with an end-point error of 0.58 pixels on the Scene Flow dataset, a D1-all error of 1.67% on the KITTI 2015 dataset, and an average disparity error of 0.22 pixels on the ETH3D dataset. Compared to other GRU-based iterative optimization algorithms, the proposed method not only exhibits significant performance advantages but also demonstrates a lightweight design.

Risk-Adjusted Returns of Private Equity Funds: A New Approach

Abstract This paper introduces a new metric, α, to benchmark the performance of individual private equity funds. Our metric is substantially less sensitive to noise in fund cash flows compared to the popular public market equivalent (PME) and its generalization (GPME), while having the same aggregate pricing implications as GPME. For a large data set of fund cash flows, α estimates have much lower standard deviation across funds than does (G)PME. For buyout funds, PME and α are close, but deviate in certain subsamples. Using α increases power in regressions involving fund performance and improves performance predictability of future funds.

Spatial–Temporal-Correlation-Constrained Dynamic Graph Convolutional Network for Traffic Flow Forecasting

Accurate traffic flow prediction in road networks is essential for intelligent transportation systems (ITS). Since traffic data are collected from the road network with spatial topological and time series sequences, the traffic flow prediction is regarded as a spatial–temporal prediction task. With the powerful ability to model the non-Euclidean data, the graph convolutional network (GCN)-based models have become the mainstream framework for traffic forecasting. However, existing GCN-based models either use the manually predefined graph structure to capture the spatial features, ignoring the heterogeneity of road networks, or simply perform 1-D convolution with fixed kernel to capture the temporal dependencies of traffic data, resulting in insufficient long-term temporal feature extraction. To solve those issues, a spatial–temporal correlation constrained dynamic graph convolutional network (STC-DGCN) is proposed for traffic flow forecasting. In STC-DGCN, a spatial–temporal embedding encoder module (STEM) is first constructed to encode the dynamic spatial relationships for road networks at different time steps. Then, a temporal feature encoder module with heterogeneous time series correlation modeling (TFE-HCM) and a spatial feature encoder module with dynamic multi-graph modeling (SFE-DCM) are designed to generate dynamic graph structures for effectively capturing the dynamic spatial and temporal correlations. Finally, a spatial–temporal feature fusion module based on a gating fusion mechanism (STM-GM) is proposed to effectively learn and leverage the inherent spatial–temporal relationships for traffic flow forecasting. Experimental results from three real-world traffic flow datasets demonstrate the superior performance of the proposed STC-DGCN compared with state-of-the-art traffic flow forecasting models.

TCANet: Three-dimensional cross-attention mechanism for stereo-matching

Abstract Effective disparity estimation is a current hotspot in stereo vision research, and cost aggregation is an important part of disparity prediction. How to perform more effective cost aggregation is the core step to improve the accuracy of disparity prediction. In previous studies, 3DCNN with a stacked hourglass structure is often used for cost aggregation. In this research, we propose an effective 3D cross-attention stereo network that utilizes the attention mechanism to obtain contextual information for cost aggregation in a more efficient way. Specifically, the 3D cross-attention module in TCANet acquires the geometric information of all pixels on the 3D cross path. By repeating this operation twice, each pixel can eventually obtain global dependencies from all other pixels. Using the 3D cross-attention module in a stacked hourglass-structured 3DCNN only increases the number of parameters by a very small amount, which can effectively improve the performance of the model. Experimental results show that TCANet performs well on virtual dataset Scene Flow and realistic KITTI datasets.

LGTCN: A Spatial–Temporal Traffic Flow Prediction Model Based on Local–Global Feature Fusion Temporal Convolutional Network

High-precision traffic flow prediction facilitates intelligent traffic control and refined management decisions. Previous research has built a variety of exquisite models with good prediction results. However, they ignore the reality that traffic flows can propagate backwards on road networks when modeling spatial relationships, as well as associations between distant nodes. In addition, more effective model components for modeling temporal relationships remain to be developed. To address the above challenges, we propose a local–global features fusion temporal convolutional network (LGTCN) for spatio-temporal traffic flow prediction, which incorporates a bidirectional graph convolutional network, probabilistic sparse self-attention, and a multichannel temporal convolutional network. To extract the bidirectional propagation relationship of traffic flow on the road network, we improve the traditional graph convolutional network so that information can be propagated in multiple directions. In addition, in spatial global dimensions, we propose probabilistic sparse self-attention to effectively perceive global data correlations and reduce the computational complexity caused by the finite perspective graph. Furthermore, we develop a multichannel temporal convolutional network. It not only retains the temporal learning capability of temporal convolutional networks, but also corresponds each channel to a node, and it realizes the interaction of node features through output interoperation. Extensive experiments on four open access benchmark traffic flow datasets demonstrate the effectiveness of our model.

Long-term Traffic Flow Prediction using Stochastic Configuration Networks for Smart Cities

Accurate predictions of traffic flow are very meaningful to city managers. With such information, traffic systems can better coordinate traffic signals and reduce congestion. By understanding traffic patterns, navigation systems can provide real-time routing suggestions that avoid traffic jams, save time, and reduce fuel consumption. However, traffic flow will be interfered with by multiple factors such as collection time and place. In this paper, we propose to use stochastic configuration networks (SCNs) to predict the traffic flow. The network is trained through stepwise construction, and the network parameters are effectively optimized based on the approximation theorem and convergence analysis optimization mechanism. The proposed network automatically adjusts its structure according to the complexity of traffic flow to better adapt to the complex non-linearity of traffic flow. We observed that the proposed model achieves better prediction performance overall and greater flexibility in the length of the prediction period compared to the benchmarks using the Guangzhou urban traffic flow dataset. It's worth noting that SCNs consistently outperform other models across different prediction intervals. They yield RMSE improvements of up to 10.73% for 10-minute predictions, 5.02% for 30-minute predictions, and 11.21% for 60-minute predictions compared to the least effective models. The R-value also exhibits steady enhancement, increasing up to 0.78%, 0.65%, and 2.33% for 10-minute, 30-minute, and 60-minute predictions, respectively. These notable advancements, combined with the model's computational efficiency, especially in short-term predictions, underscore the effectiveness and practicality of SCNs in traffic flow prediction tasks.

A Novel Prediction Model for Debris Flow Mean Velocity Based on Small Sample Data Taking Jiangjia Gully Watershed as an Example

ABSTRACTAmong all the factors affecting the destructiveness of debris flow, the mean velocity is one of the most important characteristics. In this paper, we aim to apply a particle swarm optimization (PSO) based on the relevance vector machine (RVM) to predict the mean velocity. The PSO is used to optimize kernel parameters inside the RVM, whereas the RVM is responsible for completing the prediction task. Through sample training, a nonlinear relationship can be obtained, enabling a rapid prediction of the mean velocity for new samples. The debris flow dataset of Jiangjia Gully is used to evaluate the performance of PSO‐RVM in this study. Besides, we further compare the prediction results of PSO‐RVM with other prominent approaches, for example, the support vector machine (SVM), BP neural network (BP), and the RVM. The results show that the mean relative error (MRE) of PSO‐RVM is only 0.69%. In addition, BP yields the highest MRE (27.61%), and the MRE (2.75%) corresponding to the RVM is lower than that (5.98%) yielded by the SVM. For the root mean square error (RMSE) and Theil's inequality coefficient (TIC), the PSO‐RVM method still generates much lower RMSE (6.48%) and TIC (0.179%) values than the other three methods. Overall, compared with current debris flow prediction models, the PSO‐RVM achieves high prediction accuracy, fewer optimization parameters, and low computational complexity. Finally, a sensitivity analysis is conducted to explore the dominative factors of debris flow.

Dominance of Artificial Intelligence and Machine Learning Algorithms in Real-Time Traffic Flow prediction and Route Optimization in Autonomous Vehicles

The evolution of autonomous vehicles has elicited significant interest in understanding how real-time data may be used to provide enhanced driving experiences. This research project explored AI and Machine Learning methodologies applied for traffic forecasting and route optimization, and their implications for autonomous vehicles and urban mobility. For this project, the road traffic flow Dataset was utilized from Kaggle, containing 48,000 records of the flow of traffic, each including the following key features. In our work, we deployed some credible and well-established machine learning models: linear regression and random forest. These algorithms were separately trained by using a part of preprocessed data. The MSE for the Random Forest model was significantly lower, which means that the Random Forest Regressor had much smaller errors in estimating the volume of traffic compared to the Linear Regression model. The Random Forest Model had a high R² score, proving that this model explains a great deal of variance in the volume of traffic. This means that the 'model of random forest regressors' excellently fitted the data, snatching most of the important patterns and relationships between input features and target variables.

Learning traffic as videos: A spatio-temporal VAE approach to periodic traffic raster data imputation

For modern Intelligent Transportation System (ITS), data missing during traffic raster acquisition can be inevitable because of the loop detector malfunction or signal interference. Nevertheless, missing data imputation is meaningful due to the periodic spatio-temporal characteristics and individual randomness of traffic raster data. In this paper, traffic raster data collected from all spatial regions at each time interval are considered as a multiple channel image. Accordingly, the traffic raster data over a period of time can be regarded as video, on which an unsupervised generative neural network called MSST-VAE (Multiple Streams Spatial Temporal-VAE) is proposed for traffic raster data imputation, and this model can even robustly performs at varied missing rates while many other approaches fail to conduct. Two major innovations can be summarized in MSSTVAE: Firstly, it uses multiple periodic streams of Variational Auto-Encoders (VAEs) with Sylvester Normalizing Flows (SNFs), which shows strong generalization ability. Secondly, after the traffic raster data are transferred into videos, an ECB (Extraction-and-Calibration Block) consisting of dilated P3D gated convolution and multi-horizon attention mechanism is employed to learn global-local-granularity spatial features and long-short-term temporal features. Extensive experiments on three real traffic flow datasets validate that MSST-VAE outperforms other classical traffic imputation models with the least imputation error.

4D flow MRI velocity uncertainty quantification.

An automatic method is presented for estimating 4D flow MRI velocity measurement uncertainty in each voxel. The velocity distance (VD) metric, a statistical distance between the measured velocity and local error distribution, is introduced as a novel measure of 4D flow MRI velocity measurement quality. The method uses mass conservation to assess the local velocity error variance and the standardized difference of means (SDM) velocity to estimate the velocity error correlations. VD is evaluated as the Mahalanobis distance between the local velocity measurement and the local error distribution. The uncertainty model is validated synthetically and tested in vitro under different flow resolutions and noise levels. The VD's application is demonstrated on two in vivo thoracic vasculature 4D flow datasets. Synthetic results show the proposed uncertainty quantification method is sensitive to aliased regions across various velocity-to-noise ratios and assesses velocity error correlations in four- and six-point acquisitions with correlation errors at or under 3.2%. In vitro results demonstrate the method's sensitivity to spatial resolution, venc settings, partial volume effects, and phase wrapping error sources. Applying VD to assess in vivo 4D flow MRI in the aorta demonstrates the expected increase in measured velocity quality with contrast administration and systolic flow. The proposed 4D flow MRI uncertainty quantification method assesses velocity measurement error owing to sources including noise, intravoxel phase dispersion, and velocity aliasing. This method enables rigorous comparison of 4D flow MRI datasets obtained in longitudinal studies, across patient populations, and with different MRI systems.

Privacy-preserving coordination of power and transportation networks using spatiotemporal GAT for predicting EV charging demands

The gradual replacement of conventional-fuel vehicles by electric vehicles (EVs) in recent years provides a growing incentive for the collaborative optimization of power distribution networks (PDNs) and urban transportation networks (UTNs). However, the implementation of a centralized optimization model for PDNs and UTNs is currently unrealistic due to the requirement for establishing information privacy barriers between the two independently operated networks. The present work proposes a predict-then-optimize (PTO) framework that combines Spatiotemporal Graph Attention Network (STGAT) with enhanced queueing theory to achieve accurate prediction of EV charging demands. Building upon this, it efficiently coordinates optimization between the PDN and UTN, while preserving the information privacy of both independently operated networks. Specifically, the proposed STGAT model predicts vehicle flow by simultaneously exploiting the dynamic temporal and spatial dependencies in the historical traffic data of the target UTN. It then converts the predicted flow into spatiotemporal EV charging demands using an enhanced queuing model that considers the service capacity constraints of charging stations (CSs) and the behavior of EV users. Subsequently, the predicted EV charging demands are integrated into a multi-period PDN scheduling model. The results of numerical computations based on an IEEE 33-bus PDN and real-world traffic flow datasets demonstrate that the scheduling results provided by the proposed approach differ by only 0.5% compared to results obtained when applying actual EV charging demands.

Quantifying the Complexity of Nodes in Higher-Order Networks Using the Infomap Algorithm

Accurately quantifying the complexity of nodes in a network is crucial for revealing their roles and network complexity, as well as predicting network emergent phenomena. In this paper, we propose three novel complexity metrics for nodes to reflect the extent to which they participate in organized, structured interactions in higher-order networks. Our higher-order network is built using the BuildHON+ model, where communities are detected using the Infomap algorithm. Since a physical node may contain one or more higher-order nodes in higher-order networks, it may simultaneously exist in one or more communities. The complexity of a physical node is defined by the number and size of the communities to which it belongs, as well as the number of higher-order nodes it contains within the same community. Empirical flow datasets are used to evaluate the effectiveness of the proposed metrics, and the results demonstrate their efficacy in characterizing node complexity in higher-order networks.

Purity Skeleton Dynamic Hypergraph Neural Network

Recently, in the field of Hypergraph Neural Networks (HGNNs), the effectiveness of dynamic hypergraph construction has been validated, which aims to reduce structural noise within the hypergraph through embeddings. However, the existing dynamic construction methods fail to notice the reduction of information contained in the hypergraphs during dynamic updates. This limitation undermines the quality of hypergraphs. Moreover, dynamic hypergraphs are constructed from graphs. Several key nodes play a crucial role in graph, but they are overlooked in hypergraphs. In this paper, we propose a Purity Skeleton Dynamic Hypergraph Neural Network (PS-DHGNN) to address the above issues. Firstly, we leverage purity skeleton method to dynamically construct hypergraphs via the fusion embeddings of features and topology simultaneously. This method effectively reduces structural noise and prevents the loss of information. Secondly, we employ an incremental training strategy, which implements a batch training strategy based on the importance of nodes. The key nodes, as the skeleton of hypergraph, are still highly valued. In addition, we utilize a novel loss function for learning structure information between hypergraph and graph. We conduct extensive experiments on node classification and clustering tasks, which demonstrate that our PS-DHGNN outperforms state-of-the-art methods. Note on real-world traffic flow datasets, PS-DHGNN demonstrates excellent performance, which is highly meaningful in practice.

The risk and return of impact investing funds

We provide the first analysis of the risk exposure and risk-adjusted performance of impact investing funds, private market funds with dual financial and social goals. We introduce a dataset of impact fund cash flows and exploit distortions in VC performance measures to characterize risk profiles. Impact funds have a lower market β than comparable private market strategies. Accounting for β, impact funds underperform the public market, though not necessarily more so than comparable strategies. We consider alternative pricing models, accounting for sustainability and emerging markets risk. We show investors’ wealth portfolios and taste change the perceived financial merit of impact investing.

A novel tensor-based modal decomposition method for reduced order modeling and optimal sparse sensor placement

Data-driven modal decomposition methods have become fundamental algorithms in constructing reduced order models (ROMs), which serve as a critical tool in the digital twin modeling of large complex systems. Classical modal decomposition methods are particularly adept at handling matrix-form input data that contains columnized snapshots without retaining spatial structure and neighborhood information. Focusing on addressing this problem, this paper proposed a Tensor-Based Modal Decomposition (TBMD) method to deal with the high-order tensor-form input data. TBMD method is effective in finding the potential low-order manifolds of a given high-order dynamical system with low energy loss within fewer decomposited modes. Moreover, TBMD is extended to be used in the field of optimal sensor placement via the development of a tensor-based QR factorization method and a tensor-based compressive sensing algorithm. The tensor-based QR factorization method is able to consider the spatial structure and neighborhood information of the basis matrices data in tensor form. This is the first time that a compressive sensing model has been discussed and solved for the frontal slice-wise product of a tensor and a vector. Subsequently, comprehensive case studies are conducted based on the cylinder wake dataset, airfoil vortex dataset, sea surface temperature dataset, eigenfaces dataset and turbulent channel flow dataset. The experiment results show that TBMD can accurately capture lower-order manifolds, and the tensor-based QR factorization is effective in optimizing sensor placement with fewer sensors while maintaining high reconstruction accuracy. The proposed methods in this paper is promising to be utilized in the construction of large-scale complex system digital twins, in which TBMD can serve as a fundamental algorithm in constructing ROM.

Transfer learning-based nonstationary traffic flow prediction using AdaRNN and DCORAL

Traffic flow prediction is an integral part of an intelligent transportation system (ITS) for proactive transportation planning and management in public transit network systems. However, due to the influence of passengers’ travel modes and the difficulty of data collection, some traffic flow datasets are not only nonstationary in temporal distribution but also insufficient in data size, leading to inaccurate prediction results. Therefore, a transfer learning-based method is proposed to predict nonstationary traffic flow with inadequate data in the public transit system. In the proposed model, a novel approach named AdaRNN-DCORAL is presented to achieve accurate traffic flow prediction in the temporal dimension. Specifically, AdaRNN is an adaptive prediction network for nonstationary traffic flow data, while DCORAL performs domain adaptation between the source and target domain in the public transit system. First, two domains of traffic flow datasets in two public transit modes are established, which are the target domain with insufficient data and the source domain with sufficient data. Subsequently, the traffic flow datasets are divided into weekday and holiday time series based on their distribution characteristics. On this basis, AdaRNN-DCORAL is trained on the constructed datasets, thus achieving the precise nonstationary prediction of data scarcity in the target domain. Finally, the experiments demonstrate the improved performance of the proposed method over that of state-of-the-art models in terms of prediction accuracy and computation time.

Spnaf: An R package for analyzing and mapping the hotspots of flow datasets

This paper introduces {spnaf} (spatial network autocorrelation for flows), an R package designed for the hotspot analysis of flow (e.g., human mobility, transportation, and animal movement) datasets based on Berglund and Karlström’s G index. We demonstrate the utility of the {spnaf} package through two example analyses by data forms: 1) bike-sharing trip patterns in Columbus, Ohio, USA, using polygon data, and 2) U.S. airports’ passenger travel patterns, using point data. The {spnaf} is available for download from the Comprehensive R Archive Network (CRAN), which contains a vignette and sample data/code for immediate use. This package addresses limitations in existing spatial analysis packages and emphasizes its efficiency in detecting flow hotspots. It is highly applicable in various urban and geographic data science applications. {spnaf} is still in its early stages and we hope that interested readers can contribute to the development and enhancement of the package.