OCT-GAN: Neural ODE-based Conditional Tabular GANs

Abstract

Synthesizing tabular data is attracting much attention these days for various purposes. With sophisticate synthetic data, for instance, one can augment its training data. For the past couple of years, tabular data synthesis techniques have been greatly improved. Recent work made progress to address many problems in synthesizing tabular data, such as the imbalanced distribution and multimodality problems. However, the data utility of state-of-the-art methods is not satisfactory yet. In this work, we significantly improve the utility by designing our generator and discriminator based on neural ordinary differential equations (NODEs). After showing that NODEs have theoretically preferred characteristics for generating tabular data, we introduce our designs. The NODE-based discriminator performs a hidden vector evolution trajectory-based classification rather than classifying with a hidden vector at the last layer only. Our generator also adopts an ODE layer at the very beginning of its architecture to transform its initial input vector (i.e., the concatenation of a noisy vector and a condition vector in our case) onto another latent vector space suitable for the generation process. We conduct experiments with 13 datasets, including but not limited to insurance fraud detection, online news article prediction, and so on, and our presented method outperforms other state-of-the-art tabular data synthesis methods in many cases of our classification, regression, and clustering experiments. CCS CONCEPTS• Computing methodologies → Machine learning; Neural networks.

Research on the Simulation Method of HTTP Traffic Based on GAN

“…These included CGAN(Trans), which uses the original transformer as the basic module for the generator, and CGAN(Trans-EV), which incorporates the EV algorithm. The other two models were AE-GAN, proposed by Zhu et al in 2022 [37], which employs an AE model for preprocessing data and constructs the generator using fully connected layers and batch normalization, and OCT-GAN, proposed by Kim et al in 2021 [38], which utilizes a generator built on additional ODE layers and combines them with a discriminator composed of NODEs layers. To ensure fairness in the generated data, we adopted the same GMM modeling pattern for the input data of each model, ensuring that the differences in data generation were only related to the models.…”

FinDiff: Diffusion Models for Financial Tabular Data Generation

“…To handle diverse data types more efficiently, Zhao et al (Zhao et al 2021) developed CTAB-GAN, a conditional table GAN that efficiently addresses data imbalance and distributions. Kim et al (Kim et al 2021) enhanced synthetic tabular data utility using neural ODEs, and Wen et al (Wen et al 2022) introduced Causal-TGAN, leveraging intervariable causal relationships to improve generated data quality. Zhang et al (Zhang et al 2021b) offered GANBLR for a deeper understanding of feature importance, and Noock and Guillame-Bert (Nock and Guillame-Bert 2022) proposed a tree-based approach as an interpretable alternative.…”

SOS: Score-based Oversampling for Tabular Data

“…CTGAN [44] has a conditional generator and use a mode separation process. OCT-GAN [24] exploits the homeomorphic characteristic of neural ordinary differential equations, when designing its generator, and now shows the state-of-the-art synthesis quality for many tabular datasets. However, one downside is that it requires a much longer training time than other models.…”

“…BAGAN [27] is a type of GAN model for oversampling images, but we replace its generator and discriminator with ours for tabular data synthesis. OCT-GAN [24] is one of the state of the art model for tabular data systhesis.…”