Learning Hierarchical Features from Generative Models
This addresses a bottleneck in generative modeling for researchers, offering a novel method to improve feature learning in hierarchical models.
The paper tackled the problem that hierarchical latent variable models fail to utilize their structure under existing variational training methods, and proposed an alternative architecture that learns interpretable and disentangled hierarchical features on natural image datasets without task-specific regularization.
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.