# An Architecture for Deep, Hierarchical Generative Models

arXiv: https://arxiv.org/abs/1612.04739 · HTML: https://mlsanity.com/paper/1612.04739/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 8 Dec 2016.

## Built on by (most cited first, 6 of 6)

- [Variational Predictive Routing with Nested Subjective Timescales](https://mlsanity.com/paper/2110.11236/) (2021)
- [Learning Hierarchical Features with Joint Latent Space Energy-Based Prior](https://mlsanity.com/paper/2310.09604/) (2023)
- [Kanerva++: extending The Kanerva Machine with differentiable, locally block allocated latent memory](https://mlsanity.com/paper/2103.03905/) (2021)
- [CatVRNN: Generating Category Texts via Multi-task Learning](https://mlsanity.com/paper/2107.05219/) (2021)
- [Syntax-Guided Diffusion Language Models with User-Integrated Personalization](https://mlsanity.com/paper/2510.01028/) (2025)
- [Latent Fourier Transform](https://mlsanity.com/paper/2604.17986/) (2026)

## Abstract (arXiv)

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.