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 #None
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 · Generative Models · Adversarial Networks
[ Paper] 

Author Information

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

![](https://i.imgur.com/FQwXTGR.png) [ShortScience.org profile](http://www.shortscience.org/user?name=joecohen) Joseph Paul Cohen holds a Ph.D Degree in Computer Science and Machine Learning from the University of Massachusetts Boston. His research interests include machine learning, domain adaptation, computer vision, medical applications, ad-hoc networking, and cyber security. Joseph received a U.S. National Science Foundation Graduate Fellowship in 2013 as well as COSPAR’s Outstanding Paper Award for Young Scientists in the same year. Joseph is the founder of the Institute for Reproducible Research which produces [ShortScience.org](http://shortscience.org); which lets researchers publish and read summaries of research papers like an online journal club, as well as [Academic Torrents](http://academictorrents.com); a system designed to move large datasets and become the library of the future. He is also the creator of BlindTool; a mobile application providing a sense of vision to the blind by using an artificial neural network that speaks names of objects as they are identified. Joseph is the creator of Blucat; a cross-platform Bluetooth debugging tool. He has worked in industry for small startups, large corporations, government research labs, educational museums, as well as been involved in projects sponsored by NASA and the DOE.

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

More from the Same Authors