Deep Amortized Inference for Probabilistic Programs using Adversarial Compilation

We propose an amortized inference strategy for probabilistic programs, one that learns from the past inferences to speed up the future inferences.

Our proposed inference strategy is to train neural guidance programs via a minimax game, with the probabilistic program as a correlation device. From a game-theoretical vantage point, the role of a correlation device is to enforce better outcomes by sharing information between players. The shared information, in our case, is the execution trace, which gets used for computation of payoffs in the minimax game.

Author: Mahdi Azarafrooz

The extended abstract is available at: pps18-adversarial-compilation

Poster: Adversarial Compilation

Posted in Uncategorized | Comments Off on Deep Amortized Inference for Probabilistic Programs using Adversarial Compilation

Welcome

Welcome to PPS, workshop on probabilistic programming semantics, on Tuesday, 9 January 2018, colocated right before POPL. This informal workshop aims to bring programming-language and machine-learning researchers together to advance the semantic foundations of probabilistic programming.

We are delighted that Erik Meijer has accepted our invitation to give a talk “Software is eating the world, but ML is going to eat software” and that Rif A. Saurous and Dustin Tran have accepted our invitation to give a tutorial on “Deep Probabilistic Programming: TensorFlow Distributions and Edward”. In addition, as listed on the workshop website, we have accepted 22 extended abstracts submitted by a wide range of contributors. We accepted 14 submissions as posters and 8 as talks, not on the basis of reviewer scores but based on which medium we thought would be most effective in conveying the material. So, some highly ranked submissions that are more technical in nature are accepted as posters.

To foster collaboration and establish common ground, we ask all accepted contributors to post their revised extended abstracts on this site, along with any other materials such as preprints they want to share.

Everyone is welcome to post comments, questions, and other discussion on the posts. Because probabilistic programming is a research area that bridges multiple communities with different vocabularies, comments of the flavor “I don’t understand what you mean by X” are particularly valuable!

Slides from Erik Meijer’s talk: “Software is eating the world, but ML is going to eat software”

Slides from Rif A. Saurous and Dustin Tran’s tutorial: “Deep Probabilistic Programming: TensorFlow Distributions and Edward”

Colab notebook from the tutorial

Posted in Uncategorized | Comments Off on Welcome