Learning Deep Structured Models

Liang-Chieh Chen, Alexander Schwing, Alan Yuille, Raquel Urtasun
Proceedings of the 32nd International Conference on Machine Learning, PMLR 37:1785-1794, 2015.

Abstract

Many problems in real-world applications involve predicting several random variables that are statistically related. Markov random fields (MRFs) are a great mathematical tool to encode such dependencies. The goal of this paper is to combine MRFs with deep learning to estimate complex representations while taking into account the dependencies between the output random variables. Towards this goal, we propose a training algorithm that is able to learn structured models jointly with deep features that form the MRF potentials. Our approach is efficient as it blends learning and inference and makes use of GPU acceleration. We demonstrate the effectiveness of our algorithm in the tasks of predicting words from noisy images, as well as tagging of Flickr photographs. We show that joint learning of the deep features and the MRF parameters results in significant performance gains.

Cite this Paper

BibTeX
@InProceedings{pmlr-v37-chenb15, title = {Learning Deep Structured Models}, author = {Chen, Liang-Chieh and Schwing, Alexander and Yuille, Alan and Urtasun, Raquel}, booktitle = {Proceedings of the 32nd International Conference on Machine Learning}, pages = {1785--1794}, year = {2015}, editor = {Bach, Francis and Blei, David}, volume = {37}, series = {Proceedings of Machine Learning Research}, address = {Lille, France}, month = {07--09 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v37/chenb15.pdf}, url = {https://proceedings.mlr.press/v37/chenb15.html}, abstract = {Many problems in real-world applications involve predicting several random variables that are statistically related. Markov random fields (MRFs) are a great mathematical tool to encode such dependencies. The goal of this paper is to combine MRFs with deep learning to estimate complex representations while taking into account the dependencies between the output random variables. Towards this goal, we propose a training algorithm that is able to learn structured models jointly with deep features that form the MRF potentials. Our approach is efficient as it blends learning and inference and makes use of GPU acceleration. We demonstrate the effectiveness of our algorithm in the tasks of predicting words from noisy images, as well as tagging of Flickr photographs. We show that joint learning of the deep features and the MRF parameters results in significant performance gains.} }
Endnote
%0 Conference Paper %T Learning Deep Structured Models %A Liang-Chieh Chen %A Alexander Schwing %A Alan Yuille %A Raquel Urtasun %B Proceedings of the 32nd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2015 %E Francis Bach %E David Blei %F pmlr-v37-chenb15 %I PMLR %P 1785--1794 %U https://proceedings.mlr.press/v37/chenb15.html %V 37 %X Many problems in real-world applications involve predicting several random variables that are statistically related. Markov random fields (MRFs) are a great mathematical tool to encode such dependencies. The goal of this paper is to combine MRFs with deep learning to estimate complex representations while taking into account the dependencies between the output random variables. Towards this goal, we propose a training algorithm that is able to learn structured models jointly with deep features that form the MRF potentials. Our approach is efficient as it blends learning and inference and makes use of GPU acceleration. We demonstrate the effectiveness of our algorithm in the tasks of predicting words from noisy images, as well as tagging of Flickr photographs. We show that joint learning of the deep features and the MRF parameters results in significant performance gains.
RIS
TY - CPAPER TI - Learning Deep Structured Models AU - Liang-Chieh Chen AU - Alexander Schwing AU - Alan Yuille AU - Raquel Urtasun BT - Proceedings of the 32nd International Conference on Machine Learning DA - 2015/06/01 ED - Francis Bach ED - David Blei ID - pmlr-v37-chenb15 PB - PMLR DP - Proceedings of Machine Learning Research VL - 37 SP - 1785 EP - 1794 L1 - http://proceedings.mlr.press/v37/chenb15.pdf UR - https://proceedings.mlr.press/v37/chenb15.html AB - Many problems in real-world applications involve predicting several random variables that are statistically related. Markov random fields (MRFs) are a great mathematical tool to encode such dependencies. The goal of this paper is to combine MRFs with deep learning to estimate complex representations while taking into account the dependencies between the output random variables. Towards this goal, we propose a training algorithm that is able to learn structured models jointly with deep features that form the MRF potentials. Our approach is efficient as it blends learning and inference and makes use of GPU acceleration. We demonstrate the effectiveness of our algorithm in the tasks of predicting words from noisy images, as well as tagging of Flickr photographs. We show that joint learning of the deep features and the MRF parameters results in significant performance gains. ER -
APA
Chen, L., Schwing, A., Yuille, A. & Urtasun, R.. (2015). Learning Deep Structured Models. Proceedings of the 32nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 37:1785-1794 Available from https://proceedings.mlr.press/v37/chenb15.html.

Related Material