Learning Priors for Adversarial Autoencoders
This work addresses a bottleneck in deep latent factor models for researchers and practitioners, offering incremental improvements in generative modeling tasks.
The paper tackles the problem of suboptimal priors in adversarial autoencoders by learning data-driven priors using code generators, resulting in improved image quality and better disentangled representations compared to standard AAEs.
Most deep latent factor models choose simple priors for simplicity, tractability or not knowing what prior to use. Recent studies show that the choice of the prior may have a profound effect on the expressiveness of the model,especially when its generative network has limited capacity. In this paper, we propose to learn a proper prior from data for adversarial autoencoders(AAEs). We introduce the notion of code generators to transform manually selected simple priors into ones that can better characterize the data distribution. Experimental results show that the proposed model can generate better image quality and learn better disentangled representations than AAEs in both supervised and unsupervised settings. Lastly, we present its ability to do cross-domain translation in a text-to-image synthesis task.