Timezone: »

 
Poster
GibbsNet: Iterative Adversarial Inference for Deep Graphical Models
Alex Lamb · R Devon Hjelm · Yaroslav Ganin · Joseph Paul Cohen · Aaron Courville · Yoshua Bengio

Tue Dec 05 06:30 PM -- 10:30 PM (PST) @ Pacific Ballroom #141
in Posters Tue »
Directed latent variable models that formulate the joint distribution as $p(x,z) = p(z) p(x \mid z)$ have the advantage of fast and exact sampling. However, these models have the weakness of needing to specify $p(z)$, often with a simple fixed prior that limits the expressiveness of the model. Undirected latent variable models discard the requirement that $p(z)$ be specified with a prior, yet sampling from them generally requires an iterative procedure such as blocked Gibbs-sampling that may require many steps to draw samples from the joint distribution $p(x, z)$. We propose a novel approach to learning the joint distribution between the data and a latent code which uses an adversarially learned iterative procedure to gradually refine the joint distribution, $p(x, z)$, to better match with the data distribution on each step. GibbsNet is the best of both worlds both in theory and in practice. Achieving the speed and simplicity of a directed latent variable model, it is guaranteed (assuming the adversarial game reaches the virtual training criteria global minimum) to produce samples from $p(x, z)$ with only a few sampling iterations. Achieving the expressiveness and flexibility of an undirected latent variable model, GibbsNet does away with the need for an explicit $p(z)$ and has the ability to do attribute prediction, class-conditional generation, and joint image-attribute modeling in a single model which is not trained for any of these specific tasks. We show empirically that GibbsNet is able to learn a more complex $p(z)$ and show that this leads to improved inpainting and iterative refinement of $p(x, z)$ for dozens of steps and stable generation without collapse for thousands of steps, despite being trained on only a few steps.
Keywords: Deep Learning · Adversarial Networks · Generative Models
[ Paper] 

Author Information

Alex Lamb (UMontreal (MILA))
R Devon Hjelm (MSR / MILA)
Yaroslav Ganin (DeepMind)
Joseph Paul Cohen (MILA ShortScience.org)

Joseph Paul Cohen is a researcher and pragmatic engineer. He currently focuses on the challenges in deploying AI tools in medicine specifically computer vision and genomics. He maintains many open source projects including Chester the AI radiology assistant, TorchXRayVision, and BlindTool – a mobile vision aid app. He is the director of the Institute for Reproducible Research, a US non-profit which operates ShortScience.org and Academic Torrents.

Aaron Courville (U. Montreal)
Yoshua Bengio (U. Montreal)

More from the Same Authors