Generating Multi-Categorical Samples with Generative Adversarial Networks

TL;DR

This paper introduces a novel GAN-based approach for generating multivariate categorical data, addressing challenges in modeling discrete variables with improved architectures and evaluation across diverse datasets.

Contribution

The paper presents new GAN architectures with Gumbel softmax layers tailored for multivariate categorical data, outperforming existing models.

Findings

Proposed architectures outperform existing models on various datasets.

Gumbel softmax layers effectively handle discrete categorical variables.

Method demonstrates robustness across different data sparsity and feature dependencies.

Abstract

We propose a method to train generative adversarial networks on mutivariate feature vectors representing multiple categorical values. In contrast to the continuous domain, where GAN-based methods have delivered considerable results, GANs struggle to perform equally well on discrete data. We propose and compare several architectures based on multiple (Gumbel) softmax output layers taking into account the structure of the data. We evaluate the performance of our architecture on datasets with different sparsity, number of features, ranges of categorical values, and dependencies among the features. Our proposed architecture and method outperforms existing models.