Timezone: »

Learning Hierarchical Features from Deep Generative Models
Shengjia Zhao · Jiaming Song · Stefano Ermon

Sun Aug 06 10:30 PM -- 10:48 PM (PDT) @ Parkside 1

Deep neural networks have been shown to be very successful at learning feature hierarchies in supervised learning tasks. Generative models, on the other hand, have benefited less from hierarchical models with multiple layers of latent variables. In this paper, we prove that hierarchical latent variable models do not take advantage of the hierarchical structure when trained with existing variational methods, and provide some limitations on the kind of features existing models can learn. Finally we propose an alternative architecture that do not suffer from these limitations. Our model is able to learn highly interpretable and disentangled hierarchical features on several natural image datasets with no task specific regularization or prior knowledge.

Author Information

Shengjia Zhao (Stanford University)
Jiaming Song (Stanford University)
Stefano Ermon (Stanford University)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors