Structure by Architecture: Disentangled Representations without Regularization

We study the problem of self-supervised structured representation learning using autoencoders for generative modeling. Unlike most methods which rely on matching an arbitrary, relatively unstructured, prior distribution for sampling, we propose a sampling technique that relies solely on the independence of latent variables, thereby avoiding the trade-off between reconstruction quality and generative performance inherent to VAEs. We design a novel autoencoder architecture capable of learning a structured representation without the need for aggressive regularization. Our structural decoders learn a hierarchy of latent variables, akin to structural causal models, thereby ordering the information without any additional regularization. We demonstrate how these models learn a representation that improves results in a variety of downstream tasks including generation, disentanglement, and extrapolation using several challenging and natural image datasets.