[en] 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.
Research center :
ULHPC - University of Luxembourg: High Performance Computing
Disciplines :
Computer science
Author, co-author :
Camino, Ramiro Daniel ; University of Luxembourg > Interdisciplinary Centre for Security, Reliability and Trust (SNT)
Hammerschmidt, Christian ; University of Luxembourg > Interdisciplinary Centre for Security, Reliability and Trust (SNT)
State, Radu ; University of Luxembourg > Interdisciplinary Centre for Security, Reliability and Trust (SNT)
External co-authors :
no
Language :
English
Title :
Generating Multi-Categorical Samples with Generative Adversarial Networks
Publication date :
July 2018
Event name :
ICML 2018 workshop on Theoretical Foundations and Applications of Deep Generative Models