Large mvs

Programming with Probabilistic Graphical Models

Martin Vechev

Recorded 15 December 2014 in Lausanne, Vaud, Switzerland

Event: IC Colloquia - EPFL IC School Colloquia


The increased availability of massive codebases (“Big Code”) creates an exciting opportunity for new kinds of programming tools based on probabilistic models. Enabled by these models, tomorrow’s tools will provide probabilistically likely solutions to programming tasks that are difficult or impossible to solve with traditional techniques.

In this talk, I will present a new approach for building such tools based on structured prediction with graphical models, and in particular, conditional random fields. These are powerful machine learning techniques popular in computer vision -- by connecting these techniques to programs, our work enables new applications not previously possible.

As an example, I will discuss JSNice (, a system that automatically de-minifies JavaScript programs by predicting statistically likely variable names and types. Since its release few months ago, JSNice has become a popular tool in the JavaScript community and is regularly used by thousands of developers worldwide.

Watched 1357 times.