In many applications of machine learning, labeled data is scarce and obtaining additional labels is expensive. We introduce a new approach to supervising learning algorithms without labels by enforcing a small number of domain-specific constraints over the algorithms' outputs. The constraints can be provided explicitly based on prior knowledge -- e.g. we may require that objects detected in videos satisfy the laws of physics -- or implicitly extracted from data using a novel framework inspired by adversarial training. We demonstrate the effectiveness of constraint-based learning on a variety of tasks -- including tracking, object detection, and human pose estimation -- and we find that algorithms supervised with constraints achieve high accuracies with only a small amount of labels, or with no labels at all in some cases.
%0 Journal Article
%1 RenStewartEtAll18aimag
%A Ren, Hongyu
%A Stewart, Russell
%A Song, Jiaming
%A Kuleshov, Volodymyr
%A Ermon, Stefano
%D 2018
%J AI Magazine
%K 01801 paper aaai ai learn algorithm requirements
%N 1
%P 27--38
%R 10.1609/aimag.v39i1.2776
%T Learning with Weak Supervision from Physics and Data-Driven Constraints
%V 39
%X In many applications of machine learning, labeled data is scarce and obtaining additional labels is expensive. We introduce a new approach to supervising learning algorithms without labels by enforcing a small number of domain-specific constraints over the algorithms' outputs. The constraints can be provided explicitly based on prior knowledge -- e.g. we may require that objects detected in videos satisfy the laws of physics -- or implicitly extracted from data using a novel framework inspired by adversarial training. We demonstrate the effectiveness of constraint-based learning on a variety of tasks -- including tracking, object detection, and human pose estimation -- and we find that algorithms supervised with constraints achieve high accuracies with only a small amount of labels, or with no labels at all in some cases.
@article{RenStewartEtAll18aimag,
abstract = {In many applications of machine learning, labeled data is scarce and obtaining additional labels is expensive. We introduce a new approach to supervising learning algorithms without labels by enforcing a small number of domain-specific constraints over the algorithms' outputs. The constraints can be provided explicitly based on prior knowledge -- e.g. we may require that objects detected in videos satisfy the laws of physics -- or implicitly extracted from data using a novel framework inspired by adversarial training. We demonstrate the effectiveness of constraint-based learning on a variety of tasks -- including tracking, object detection, and human pose estimation -- and we find that algorithms supervised with constraints achieve high accuracies with only a small amount of labels, or with no labels at all in some cases.},
added-at = {2018-04-17T15:22:20.000+0200},
author = {Ren, Hongyu and Stewart, Russell and Song, Jiaming and Kuleshov, Volodymyr and Ermon, Stefano},
biburl = {https://www.bibsonomy.org/bibtex/230fb7466125fc16bc1278b6d1fb9dd8c/flint63},
doi = {10.1609/aimag.v39i1.2776},
file = {AAAI online:2018/RenStewartEtAll18aimag.pdf:PDF},
groups = {public},
interhash = {e306b31a324d15211bfcabe92302231a},
intrahash = {30fb7466125fc16bc1278b6d1fb9dd8c},
issn = {0738-4602},
journal = {AI Magazine},
keywords = {01801 paper aaai ai learn algorithm requirements},
month = {#mar#},
number = 1,
pages = {27--38},
timestamp = {2018-04-17T15:22:20.000+0200},
title = {Learning with Weak Supervision from Physics and Data-Driven Constraints},
username = {flint63},
volume = 39,
year = 2018
}