An Architecture for Deep, Hierarchical Generative Models
This work addresses the challenge of scaling deep generative models for image generation and analysis, though it appears incremental in its architectural improvements.
The authors tackled the problem of training deep generative models with many latent layers by introducing an architecture with deterministic paths and richer connections, achieving state-of-the-art performance on image modeling benchmarks.
We present an architecture which lets us train deep, directed generative models with many layers of latent variables. We include deterministic paths between all latent variables and the generated output, and provide a richer set of connections between computations for inference and generation, which enables more effective communication of information throughout the model during training. To improve performance on natural images, we incorporate a lightweight autoregressive model in the reconstruction distribution. These techniques permit end-to-end training of models with 10+ layers of latent variables. Experiments show that our approach achieves state-of-the-art performance on standard image modelling benchmarks, can expose latent class structure in the absence of label information, and can provide convincing imputations of occluded regions in natural images.