Composing Graphical Models with Neural Networks for Structured Representations and Fast Inference

Johnson, M. J., Duvenaud, D., Wiltschko, A. B., Datta, S. R., & Adams, R. P. (2016). Composing Graphical Models with Neural Networks for Structured Representations and Fast Inference. Advances in Neural Information Processing Systems (NIPS) 29.
We propose a general modeling and inference framework that composes probabilistic graphical models with deep learning methods and combines their respective strengths. Our model family augments graphical structure in latent variables with neural network observation models. For inference, we extend variational autoencoders to use graphical model approximating distributions with recognition networks that output conjugate potentials. All components of these models are learned simultaneously with a single objective, giving a scalable algorithm that leverages stochastic variational inference, natural gradients, graphical model message passing, and the reparameterization trick. We illustrate this framework with several example models and an application to mouse behavioral phenotyping.
  @conference{johnson2016svae,
  year = {2016},
  author = {Johnson, Matthew J. and Duvenaud, David and Wiltschko, Alexander B. and Datta, Sandeep Robert and Adams, Ryan P.},
  title = {Composing Graphical Models with Neural Networks for Structured Representations and Fast Inference},
  booktitle = {Advances in Neural Information Processing Systems (NIPS) 29},
  keywords = {NIPS, deep learning, variational inference, graphical models},
  note = {arXiv:1603.06277 [stat.ML]}
}