Mon 24 Jun 2019 09:05 - 09:25 at 224AB - Language Design I Chair(s): Benjamin Zorn

We propose a new probabilistic programming language for the design and analysis of perception systems, especially those based on machine learning. Specifically, we consider the problems of training a perception system to handle rare events, testing its performance under different conditions, and debugging failures. We show how a probabilistic programming language can help address these problems by specifying distributions encoding interesting types of inputs and sampling these to generate specialized training and test sets. More generally, such languages can be used for cyber-physical systems and robotics to write environment models, an essential prerequisite to any formal analysis. In this paper, we focus on systems like autonomous cars and robots, whose environment is a scene, a configuration of physical objects and agents. We design a domain-specific language, Scenic, for describing scenarios that are distributions over scenes. As a probabilistic programming language, Scenic allows assigning distributions to features of the scene, as well as declaratively imposing hard and soft constraints over the scene. We develop specialized techniques for sampling from the resulting distribution, taking advantage of the structure provided by Scenic's domain-specific syntax. Finally, we apply Scenic in a case study on a convolutional neural network designed to detect cars in road images, improving its performance beyond that achieved by state-of-the-art synthetic data generation methods.

Mon 24 Jun

Displayed time zone: Tijuana, Baja California change

08:45 - 09:45
Language Design IPLDI Research Papers at 224AB
Chair(s): Benjamin Zorn Microsoft Research
08:45
20m
Talk
LoCal: A Language for Programs Operating on Serialized Data
PLDI Research Papers
Michael Vollmer Indiana University, USA, Chaitanya Koparkar Indiana University, Mike Rainey Indiana University, USA, Laith Sakka Purdue University, Milind Kulkarni Purdue University, Ryan R. Newton Indiana University
DOI Authorizer link Pre-print Media Attached
09:05
20m
Talk
Scenic: A Language for Scenario Specification and Scene Generation
PLDI Research Papers
Daniel J. Fremont University of California at Berkeley, USA, Tommaso Dreossi University of California at Berkeley, Shromona Ghosh University of California at Berkeley, USA, Xiangyu Yue University of California at Berkeley, USA, Alberto L. Sangiovanni-Vincentelli University of California at Berkeley, USA, Sanjit Seshia UC Berkeley
Media Attached
09:25
20m
Talk
Compiling KB-Sized Machine Learning Models to Tiny IoT Devices
PLDI Research Papers
Sridhar Gopinath Microsoft Research, India, Nikhil Ghanathe Microsoft Research, India, Vivek Seshadri Microsoft Research, India, Rahul Sharma Microsoft Research
Link to publication DOI Media Attached