Abstract:Some recent studies have suggested using GANs for numeric data generation such as to generate data for completing the imbalanced numeric data. Considering the significant difference between the dimensions of the numeric data and images, as well as the strong correlations between features of numeric data, the conventional GANs normally face an overfitting problem, consequently leads to an ill-conditioning problem in generating numeric and structured data. This paper studies the constrained network structures between generator G and discriminator D in WGAN, designs several structures including isomorphic, mirror and self-symmetric structures. We evaluates the performances of the constrained WGANs in data augmentations, taking the non-constrained GANs and WGANs as the baselines. Experiments prove the constrained structures have been improved in 17/20 groups of experiments. In twenty experiments on four UCI Machine Learning Repository datasets, Australian Credit Approval data, German Credit data, Pima Indians Diabetes data and SPECT heart data facing five conventional classifiers. Especially, Isomorphic WGAN is the best in 15/20 experiments. Finally, we theoretically proves that the effectiveness of constrained structures by the directed graphic model (DGM) analysis.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are: when generating numeric data, the over - fitting problem and ill - conditioning problem faced by traditional Generative Adversarial Networks (GANs). These problems lead to the low quality of the generated numeric data, especially performing poorly when dealing with unbalanced datasets. Specifically: 1. **Differences between numeric data and image data**: - Numeric data usually has a lower dimension. For example, the Pima Indians Diabetes dataset has 8 dimensions, and the SPECT Heart dataset has 22 dimensions. - The correlation between each dimension of numeric data is relatively strong, and the value of each dimension usually has a specific meaning (such as age, income, etc.), while the pixel values of image data usually do not have a clear practical meaning. 2. **Deficiencies of traditional GAN**: - Due to the above characteristics, traditional GANs are prone to over - fitting problems when generating numeric data, resulting in poor - quality generated data, which further affects the performance of classifiers. - Compared with traditional data generation methods (such as SMOTE), GAN - based methods do not show obvious advantages on some datasets. 3. **Proposed solutions**: - By introducing constrained network structures, the paper designs several WGANs (Wasserstein GAN) with specific structures, including isomorphic structure, mirror structure, and self - symmetric structure, to improve the quality of generated numeric data. - These structures aim to provide additional constraints for the learning processes of the generator G and the discriminator D, thereby improving the quality of generated data and the performance of classifiers. In summary, the main goal of this paper is to solve the over - fitting and ill - conditioning problems encountered when generating numeric data by improving the network structure of WGAN, thereby improving the quality of generated data and the performance of classifiers.

Study of Constrained Network Structures for WGANs on Numeric Data Generation

Structure-preserving GANs

Some Theoretical Insights into Wasserstein GANs

Densely Stacked Generative Adversarial Networks

Dist-GAN: An Improved GAN using Distance Constraints

Distribution Enhancement for Imbalanced Data with Generative Adversarial Network

Structure and Illumination Constrained GAN for Medical Image Enhancement

Gene-CWGAN: a data enhancement method for gene expression profile based on improved CWGAN-GP

Towards the Gradient Vanishing, Divergence Mismatching and Mode Collapse of Generative Adversarial Nets

Mdwgan-gp: data augmentation for gene expression data based on multiple discriminator WGAN-GP

Structured Generative Adversarial Networks.

CHAIN: Enhancing Generalization in Data-Efficient GANs via lipsCHitz continuity constrAIned Normalization

On the Anomalous Generalization of GANs

Inferential Wasserstein generative adversarial networks

Gang of GANs: Generative Adversarial Networks with Maximum Margin Ranking

On the Statistical Properties of Generative Adversarial Models for Low Intrinsic Data Dimension

Graphical Generative Adversarial Networks.

Data-Efficient GAN Training Beyond (Just) Augmentations: A Lottery Ticket Perspective

Hausdorff GAN: Improving GAN Generation Quality With Hausdorff Metric

Improving GAN Training via Feature Space Shrinkage