Multimodal Generative Learning Utilizing Jensen-Shannon-Divergence

Neural Information Processing Systems 

Learning from different data types is a long-standing goal in machine learning research, as multiple information sources co-occur when describing natural phenomena.