T-GCN: A Temporal Graph Convolutional Network for Traffic Prediction

Zhao, Ling; Song, Yujiao; Zhang, Chao; Liu, Yu; Wang, Pu; Lin, Tao; Deng, Min; Li, Haifeng

doi:10.1109/tits.2019.2935152

Cited by 1,926 publications

(916 citation statements)

References 33 publications

Supporting

Mentioning

710

Contrasting

Unclassified

Order By: Relevance

“…At the same time, we will test new continual learning methods using the CLRS as the baseline for developing state-of-the-art algorithms in the field of remote sensing image scene classification. We will also extend the continual learning to other geo-spatial field such as graph convolutional networks on traffic predication [40] and geo big data analysis [41].…”

Section: Discussionmentioning

confidence: 99%

CLRS: Continual Learning Benchmark for Remote Sensing Image Scene Classification

Jiang

et al. 2020

Sensors

Self Cite

View full text Add to dashboard Cite

Remote sensing image scene classification has a high application value in the agricultural, military, as well as other fields. A large amount of remote sensing data is obtained every day. After learning the new batch data, scene classification algorithms based on deep learning face the problem of catastrophic forgetting, that is, they cannot maintain the performance of the old batch data. Therefore, it has become more and more important to ensure that the scene classification model has the ability of continual learning, that is, to learn new batch data without forgetting the performance of the old batch data. However, the existing remote sensing image scene classification datasets all use static benchmarks and lack the standard to divide the datasets into a number of sequential learning training batches, which largely limits the development of continual learning in remote sensing image scene classification. First, this study gives the criteria for training batches that have been partitioned into three continual learning scenarios, and proposes a large-scale remote sensing image scene classification database called the Continual Learning Benchmark for Remote Sensing (CLRS). The goal of CLRS is to help develop state-of-the-art continual learning algorithms in the field of remote sensing image scene classification. In addition, in this paper, a new method of constructing a large-scale remote sensing image classification database based on the target detection pretrained model is proposed, which can effectively reduce manual annotations. Finally, several mainstream continual learning methods are tested and analyzed under three continual learning scenarios, and the results can be used as a baseline for future work.

show abstract

Section: Discussionmentioning

confidence: 99%

CLRS: Continual Learning Benchmark for Remote Sensing Image Scene Classification

Jiang

et al. 2020

Sensors

Self Cite

View full text Add to dashboard Cite

show abstract

“…In future study, we will try to improve the geospatial data distribution in the computing cluster to break through the I/O bottleneck. We also will improve our method by hiring technologies such as temporal graph convolutional networks [32] and deep inferring network technologies [33] to predict the performance bottleneck in applications.…”

Section: Discussionmentioning

confidence: 99%

Parallel Spatial-Data Conversion Engine: Enabling Fast Sharing of Massive Geospatial Data

Zhang

Chen

et al. 2020

Symmetry

View full text Add to dashboard Cite

Large-scale geospatial data have accumulated worldwide in the past decades. However, various data formats often result in a geospatial data sharing problem in the geographical information system community. Despite the various methodologies proposed in the past, geospatial data conversion has always served as a fundamental and efficient way of sharing geospatial data. However, these methodologies are beginning to fail as data increase. This study proposes a parallel spatial data conversion engine (PSCE) with a symmetric mechanism to achieve the efficient sharing of massive geodata by utilizing high-performance computing technology. This engine is designed in an extendable and flexible framework and can customize methods of reading and writing particular spatial data formats. A dynamic task scheduling strategy based on the feature computing index is introduced in the framework to improve load balancing and performance. An experiment is performed to validate the engine framework and performance. In this experiment, geospatial data are stored in the vector spatial data defined in the Chinese Geospatial Data Transfer Format Standard in a parallel file system (Lustre Cluster). Results show that the PSCE has a reliable architecture that can quickly cope with massive spatial datasets. . Moreover, semantic heterogeneity often occurs at the information level, requiring systems to understand the content of an item of information and its meaning, that is, semantics [9,10].Despite such obstacles, the reusability of existing data is still essential simply because of the high cost of acquiring new geographical data from scratch. Many researchers have presented various approaches that enable spatial data sharing and example systems [6,[8][9][10][11][12][13][14][15][16][17][18][19][20]. These approaches may be classified into three general categories, namely, spatial data conversion, spatial federated database, and mediator-based data sharing.Spatial data conversion techniques are the basic methods for spatial data sharing. They can be traced to the beginning of the GIS industry when a huge amount of spatial data were collected in isolated systems or departments, and the need for data sharing increased among geospatial data producers. People have attempted to convert spatial data from data sources directly into their own systems to overcome data sharing problems [8,9,14]. Currently, spatial data conversion techniques have developed extremely well and gained the power to transfer between two formats in over a hundred cases. However, such techniques often cause data redundancy and result in improper stress when coping with large datasets.Federated database techniques soon took the focus because they had several advantages over data conversion methodologies. They have allowed each database owner to define the subsets of local data and then integrate them into one virtual database by establishing a global schema in a common data model, thereby allowing users to access the federated database like a centralized database, without having...

show abstract

“…Moreover, continuing this research we would like to try other units for data aggregation-such as a square area within the city (for instance, 3 × 3 m, 10 × 10, or similar), instead of the street segment that we were using for this current research. As per the most recent research in machine learning/deep learning and urban science [42,[47][48][49][50][51], we assume that using a square area of small granularity will bring our prediction closer to the exact location where and when a crime can happen.…”

Section: Discussionmentioning

confidence: 99%

Predicting Safe Parking Spaces: A Machine Learning Approach to Geospatial Urban and Crime Data

2019

View full text Add to dashboard Cite

This research aims to identify spatial and time patterns of theft in Manhattan, NY, to reveal urban factors that contribute to thefts from motor vehicles and to build a prediction model for thefts. Methods include time series and hot spot analysis, linear regression, elastic-net, Support vector machines SVM with radial and linear kernels, decision tree, bagged CART, random forest, and stochastic gradient boosting. Machine learning methods reveal that linear models perform better on our data (linear regression, elastic-net), specifying that a higher number of subway entrances, graffiti, and restaurants on streets contribute to higher theft rates from motor vehicles. Although the prediction model for thefts meets almost all assumptions (five of six), its accuracy is 77%, suggesting that there are other undiscovered factors making a contribution to the generation of thefts. As an output demonstrating final results, the application prototype for searching safer parking in Manhattan, NY based on the prediction model, has been developed.

show abstract

T-GCN: A Temporal Graph Convolutional Network for Traffic Prediction

Cited by 1,926 publications

References 33 publications

CLRS: Continual Learning Benchmark for Remote Sensing Image Scene Classification

CLRS: Continual Learning Benchmark for Remote Sensing Image Scene Classification

Parallel Spatial-Data Conversion Engine: Enabling Fast Sharing of Massive Geospatial Data

Predicting Safe Parking Spaces: A Machine Learning Approach to Geospatial Urban and Crime Data

Contact Info

Product

Resources

About