Research Area:  Machine Learning
Learning from different data types is a long-standing goal in machine learning research, as multiple information sources co-occur when describing natural phenomena. However, existing generative models that approximate a multimodal ELBO rely on difficult or inefficient training schemes to learn a joint distribution and the dependencies between modalities. In this work, we propose a novel, efficient objective function that utilizes the Jensen-Shannon divergence for multiple distributions. It simultaneously approximates the unimodal and joint multimodal posteriors directly via a dynamic prior. In addition, we theoretically prove that the new multimodal JS-divergence (mmJSD) objective optimizes an ELBO. In extensive experiments, we demonstrate the advantage of the proposed mmJSD model compared to previous work in unsupervised, generative learning tasks.
Keywords:  
Multimodal
Generative Learning
JS-divergence
unsupervised
Machine learning
Author(s) Name:  Thomas M. Sutter, Imant Daunhawer, Julia E. Vogt
Journal name:  Machine Learning
Conferrence name:  
Publisher name:  arXiv:2006.08242
DOI:  10.48550/arXiv.2006.08242
Volume Information:  
Paper Link:   https://arxiv.org/abs/2006.08242