How Hard is Inference for Structured Prediction?

Amir Globerson, Tim Roughgarden, David Sontag, Cafer Yildirim
Proceedings of the 32nd International Conference on Machine Learning, PMLR 37:2181-2190, 2015.

Abstract

Structured prediction tasks in machine learning involve the simultaneous prediction of multiple labels. This is often done by maximizing a score function on the space of labels, which decomposes as a sum of pairwise elements, each depending on two specific labels. The goal of this paper is to develop a theoretical explanation of the empirical effectiveness of heuristic inference algorithms for solving such structured prediction problems. We study the minimum-achievable expected Hamming error in such problems, highlighting the case of 2D grid graphs, which are common in machine vision applications. Our main theorems provide tight upper and lower bounds on this error, as well as a polynomial-time algorithm that achieves the bound.

Cite this Paper

BibTeX
@InProceedings{pmlr-v37-globerson15, title = {How Hard is Inference for Structured Prediction?}, author = {Globerson, Amir and Roughgarden, Tim and Sontag, David and Yildirim, Cafer}, booktitle = {Proceedings of the 32nd International Conference on Machine Learning}, pages = {2181--2190}, 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/globerson15.pdf}, url = {https://proceedings.mlr.press/v37/globerson15.html}, abstract = {Structured prediction tasks in machine learning involve the simultaneous prediction of multiple labels. This is often done by maximizing a score function on the space of labels, which decomposes as a sum of pairwise elements, each depending on two specific labels. The goal of this paper is to develop a theoretical explanation of the empirical effectiveness of heuristic inference algorithms for solving such structured prediction problems. We study the minimum-achievable expected Hamming error in such problems, highlighting the case of 2D grid graphs, which are common in machine vision applications. Our main theorems provide tight upper and lower bounds on this error, as well as a polynomial-time algorithm that achieves the bound.} }
Endnote
%0 Conference Paper %T How Hard is Inference for Structured Prediction? %A Amir Globerson %A Tim Roughgarden %A David Sontag %A Cafer Yildirim %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-globerson15 %I PMLR %P 2181--2190 %U https://proceedings.mlr.press/v37/globerson15.html %V 37 %X Structured prediction tasks in machine learning involve the simultaneous prediction of multiple labels. This is often done by maximizing a score function on the space of labels, which decomposes as a sum of pairwise elements, each depending on two specific labels. The goal of this paper is to develop a theoretical explanation of the empirical effectiveness of heuristic inference algorithms for solving such structured prediction problems. We study the minimum-achievable expected Hamming error in such problems, highlighting the case of 2D grid graphs, which are common in machine vision applications. Our main theorems provide tight upper and lower bounds on this error, as well as a polynomial-time algorithm that achieves the bound.
RIS
TY - CPAPER TI - How Hard is Inference for Structured Prediction? AU - Amir Globerson AU - Tim Roughgarden AU - David Sontag AU - Cafer Yildirim BT - Proceedings of the 32nd International Conference on Machine Learning DA - 2015/06/01 ED - Francis Bach ED - David Blei ID - pmlr-v37-globerson15 PB - PMLR DP - Proceedings of Machine Learning Research VL - 37 SP - 2181 EP - 2190 L1 - http://proceedings.mlr.press/v37/globerson15.pdf UR - https://proceedings.mlr.press/v37/globerson15.html AB - Structured prediction tasks in machine learning involve the simultaneous prediction of multiple labels. This is often done by maximizing a score function on the space of labels, which decomposes as a sum of pairwise elements, each depending on two specific labels. The goal of this paper is to develop a theoretical explanation of the empirical effectiveness of heuristic inference algorithms for solving such structured prediction problems. We study the minimum-achievable expected Hamming error in such problems, highlighting the case of 2D grid graphs, which are common in machine vision applications. Our main theorems provide tight upper and lower bounds on this error, as well as a polynomial-time algorithm that achieves the bound. ER -
APA
Globerson, A., Roughgarden, T., Sontag, D. & Yildirim, C.. (2015). How Hard is Inference for Structured Prediction?. Proceedings of the 32nd International Conference on Machine Learning, in Proceedings of Machine Learning Research 37:2181-2190 Available from https://proceedings.mlr.press/v37/globerson15.html.

Related Material