Wei Wang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Wei Wang has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
Ruoyu Ma
Department of Computer and Control Engineering, Nankai University, Tianjin 300071, China
3 shared publicationsCesunica E. Ivey
Department of Physics, University of Nevada Reno, Reno, Nevada, USA 89557
2 shared publications- Guoliang Shi
Nankai University
2 shared publications Jian Gao
Chinese Research Academy of Environmental Sciences, Beijing 100012, China
2 shared publicationsJiayuan Liu
State Environmental Protection Key Laboratory of Urban Ambient Air Particulate Matter Pollution Prevention and Control & Center for Urban Transport Emission Research, College of Environmental Science and Engineering, Nankai University, Tianjin 300350, China
2 shared publicationsXing Peng
State Environmental Protection Key Laboratory of Urban Ambient Air Particulate Matter Pollution Prevention and Control & Center for Urban Transport Emission Research, College of Environmental Science and Engineering, Nankai University, Tianjin 300350, China
2 shared publications
Latest Publications
- Open Access
- 11 May 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 89812-89821
Study of Restrained Network Structures for Wasserstein Generative Adversarial Networks (WGANs) on Numeric Data Augmentation
Some recent studies have suggested using Generative Adversarial Network (GAN) for numeric data over-sampling, which is to generate data for completing the imbalanced numeric data. Compared with the conventional over-sampling methods, taken SMOTE as an example, the recently-proposed GAN schemes fail to generate distinguishable augmentation results for classifiers. In this paper, we discuss the reason for such failures, based on which we further study the restrained conditions between G and D theoretically, and propose a quantitative indicator of the restrained structure, called Similarity of the Restrained Condition (SRC) to measure the restrained conditions. Practically, we propose several candidate solutions, which are isomorphic (IWGAN) mirror (MWGAN) and self-symmetric WGAN (SWGAN) for restrained conditions. Besides, the restrained WGANs enhance the classification performance in AUC on five classifiers compared with the original data as the baseline, conventional SMOTE, and other GANs add up to 20 groups of experiments in four datasets. The restrained WGANs outperform all others in 17/20 groups, among which IWGAN accounted for 15/17 groups and the SRC is an effective measure in evaluating the restraints so that further GAN structures with G-D restrains could be designed on SRC. Multidimensional scaling (MDS) is introduced to eliminate the impact of datasets and evaluation of the AUC in a composite index and IWGAN decreases the MDS distance by 20% to 40%. Moreover, the convergence speed of IWGAN is increased, and the initial error of loss function is reduced.
- 11 May 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 8, 89812-89821
Open AccessStudy of Restrained Network Structures for Wasserstein Generative Adversarial Networks (WGANs) on Numeric Data Augmentation
Some recent studies have suggested using Generative Adversarial Network (GAN) for numeric data over-sampling, which is to generate data for completing the imbalanced numeric data. Compared with the conventional over-sampling methods, taken SMOTE as an example, the recently-proposed GAN schemes fail to generate distinguishable augmentation results for classifiers. In this paper, we discuss the reason for such failures, based on which we further study the restrained conditions between G and D theoretically, and propose a quantitative indicator of the restrained structure, called Similarity of the Restrained Condition (SRC) to measure the restrained conditions. Practically, we propose several candidate solutions, which are isomorphic (IWGAN) mirror (MWGAN) and self-symmetric WGAN (SWGAN) for restrained conditions. Besides, the restrained WGANs enhance the classification performance in AUC on five classifiers compared with the original data as the baseline, conventional SMOTE, and other GANs add up to 20 groups of experiments in four datasets. The restrained WGANs outperform all others in 17/20 groups, among which IWGAN accounted for 15/17 groups and the SRC is an effective measure in evaluating the restraints so that further GAN structures with G-D restrains could be designed on SRC. Multidimensional scaling (MDS) is introduced to eliminate the impact of datasets and evaluation of the AUC in a composite index and IWGAN decreases the MDS distance by 20% to 40%. Moreover, the convergence speed of IWGAN is increased, and the initial error of loss function is reduced.
- 1 May 2018
- Published by Elsevier in Environmental Pollution
- Vol. 236, 366-372
Quadratic discriminant analysis model for assessing the risk of cadmium pollution for paddy fields in a county in China
- 1 May 2018
- Published by Elsevier in Environmental Pollution
- Vol. 236, 366-372
Quadratic discriminant analysis model for assessing the risk of cadmium pollution for paddy fields in a county in China
- Open Access
- 7 March 2018
- Published by Copernicus GmbH in EGUsphere
- Vol. 2018, 1-40
A Preliminary Assessment of the Impacts of Multiple Temporal-scale Variations in Particulate Matter on its Source Apportionment
Time series of pollutant concentrations consist of variations at different time scales that are attributable to many processes/sources (data noise, source intensities, meteorological conditions, climate, etc.). Improving the knowledge of the impact of multiple temporal-scale components on pollutant variations and pollution levels can provide useful information for suitable mitigation strategies for pollutant control during a high pollution episode. To investigate the source factors driving these variations, the Kolmogorov-Zurbenko (KZ) filter was used to decompose the time series of PM2.5 (particulate matter with an aerodynamic diameter less than 2.5 μm) and chemical species into intra-day, diurnal, synoptic, and baseline temporal-scale components (TS components). The synoptic TS component has the largest amplitude and relative contributions (about 50 %) to the total variance of SO42−, NH4+, and OC concentrations. The diurnal TS component has the largest relative contributions to the total variance of PM2.5, NO3−, EC, Ca, and Fe concentrations, ranging from 32 % to 47 %. To investigate the source impacts on PM2.5 from different TS components, four datasets RI (intra-day removed), RD (diurnal removed), RS (synoptic removed), and RBL (baseline removed) were created by respectively removing the intra-day, diurnal, synoptic, and baseline TS component from the original datasets. Multilinear Engine 2 (ME-2) and/or principal component analysis was applied to these four datasets as well as the original datasets for source apportionment. ME-2 solutions using the original and RI dataset identify crustal dust contributions. For the solutions from original, RI, RD, and RS datasets, the total primary source impacts are close, ranging from 35.1 to 40.4 μg m−3 during the entire sampling period. For the secondary source impacts, solutions from the original, RI and RD dataset give similar source impacts (about 30 μg m−3), which were higher than the impacts derived from the RS datasets (21.2 μg m−3).
- 7 March 2018
- Published by Copernicus GmbH in EGUsphere
- Vol. 2018, 1-40
Open AccessA Preliminary Assessment of the Impacts of Multiple Temporal-scale Variations in Particulate Matter on its Source Apportionment
Time series of pollutant concentrations consist of variations at different time scales that are attributable to many processes/sources (data noise, source intensities, meteorological conditions, climate, etc.). Improving the knowledge of the impact of multiple temporal-scale components on pollutant variations and pollution levels can provide useful information for suitable mitigation strategies for pollutant control during a high pollution episode. To investigate the source factors driving these variations, the Kolmogorov-Zurbenko (KZ) filter was used to decompose the time series of PM2.5 (particulate matter with an aerodynamic diameter less than 2.5 μm) and chemical species into intra-day, diurnal, synoptic, and baseline temporal-scale components (TS components). The synoptic TS component has the largest amplitude and relative contributions (about 50 %) to the total variance of SO42−, NH4+, and OC concentrations. The diurnal TS component has the largest relative contributions to the total variance of PM2.5, NO3−, EC, Ca, and Fe concentrations, ranging from 32 % to 47 %. To investigate the source impacts on PM2.5 from different TS components, four datasets RI (intra-day removed), RD (diurnal removed), RS (synoptic removed), and RBL (baseline removed) were created by respectively removing the intra-day, diurnal, synoptic, and baseline TS component from the original datasets. Multilinear Engine 2 (ME-2) and/or principal component analysis was applied to these four datasets as well as the original datasets for source apportionment. ME-2 solutions using the original and RI dataset identify crustal dust contributions. For the solutions from original, RI, RD, and RS datasets, the total primary source impacts are close, ranging from 35.1 to 40.4 μg m−3 during the entire sampling period. For the secondary source impacts, solutions from the original, RI and RD dataset give similar source impacts (about 30 μg m−3), which were higher than the impacts derived from the RS datasets (21.2 μg m−3).