Minimizing Expected Losses in Perturbation Models with Multidimensional Parametric Min-cuts

Adrian Kim Seoul National University, Kyomin Jung Daniel Tarlow Pushmeet Kohli
Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence, PMLR R13:960-968, 2015.

Abstract

We consider the problem of learning perturbation-based probabilistic models by computing and differentiating expected losses. This is a challenging computational problem that has traditionally been tackled using Monte Carlo-based methods. In this work, we show how a generalization of parametric min-cuts can be used to address the same problem, achieving high accuracy of faster than a sampling-based baseline. Utilizing our proposed \textit{Skeleton Method}, we show that we can learn the perturbation model so as to directly minimize expected losses. Experimental results show that this approach offers promise as a new way of training structured prediction models under complex loss functions.

Cite this Paper

BibTeX
@InProceedings{pmlr-vR13-university15w, title = {Minimizing Expected Losses in Perturbation Models with Multidimensional Parametric Min-cuts}, author = {University, Adrian Kim Seoul National and Kohli, Kyomin Jung Daniel Tarlow Pushmeet}, booktitle = {Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence}, pages = {960--968}, year = {2015}, editor = {Meila, Marina and Heskes, Tom}, volume = {R13}, series = {Proceedings of Machine Learning Research}, month = {12--16 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/r13/main/assets/university15w/university15w.pdf}, url = {https://proceedings.mlr.press/r13/university15w.html}, abstract = {We consider the problem of learning perturbation-based probabilistic models by computing and differentiating expected losses. This is a challenging computational problem that has traditionally been tackled using Monte Carlo-based methods. In this work, we show how a generalization of parametric min-cuts can be used to address the same problem, achieving high accuracy of faster than a sampling-based baseline. Utilizing our proposed \textit{Skeleton Method}, we show that we can learn the perturbation model so as to directly minimize expected losses. Experimental results show that this approach offers promise as a new way of training structured prediction models under complex loss functions.}, note = {Reissued by PMLR on 04 October 2026.} }
Endnote
%0 Conference Paper %T Minimizing Expected Losses in Perturbation Models with Multidimensional Parametric Min-cuts %A Adrian Kim Seoul National University %A Kyomin Jung Daniel Tarlow Pushmeet Kohli %B Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence %C Proceedings of Machine Learning Research %D 2015 %E Marina Meila %E Tom Heskes %F pmlr-vR13-university15w %I PMLR %P 960--968 %U https://proceedings.mlr.press/r13/university15w.html %V R13 %X We consider the problem of learning perturbation-based probabilistic models by computing and differentiating expected losses. This is a challenging computational problem that has traditionally been tackled using Monte Carlo-based methods. In this work, we show how a generalization of parametric min-cuts can be used to address the same problem, achieving high accuracy of faster than a sampling-based baseline. Utilizing our proposed \textit{Skeleton Method}, we show that we can learn the perturbation model so as to directly minimize expected losses. Experimental results show that this approach offers promise as a new way of training structured prediction models under complex loss functions. %Z Reissued by PMLR on 04 October 2026.
APA
University, A.K.S.N. & Kohli, K.J.D.T.P.. (2015). Minimizing Expected Losses in Perturbation Models with Multidimensional Parametric Min-cuts. Proceedings of the 31st Conference on Uncertainty in Artificial Intelligence, in Proceedings of Machine Learning Research R13:960-968 Available from https://proceedings.mlr.press/r13/university15w.html. Reissued by PMLR on 04 October 2026.

Related Material