Energy-Based Processes for Exchangeable Data

Mengjiao Yang, Bo Dai, Hanjun Dai, Dale Schuurmans
Proceedings of the 37th International Conference on Machine Learning, PMLR 119:10681-10692, 2020.

Abstract

Recently there has been growing interest in modeling sets with exchangeability such as point clouds. A shortcoming of current approaches is that they restrict the cardinality of the sets considered or can only express limited forms of distribution over unobserved data. To overcome these limitations, we introduce Energy-Based Processes (EBPs), which extend energy based models to exchangeable data while allowing neural network parameterizations of the energy function. A key advantage of these models is the ability to express more flexible distributions over sets without restricting their cardinality. We develop an efficient training procedure for EBPs that demonstrates state-of-the-art performance on a variety of tasks such as point cloud generation, classification, denoising, and image completion

Cite this Paper

BibTeX
@InProceedings{pmlr-v119-yang20b, title = {Energy-Based Processes for Exchangeable Data}, author = {Yang, Mengjiao and Dai, Bo and Dai, Hanjun and Schuurmans, Dale}, booktitle = {Proceedings of the 37th International Conference on Machine Learning}, pages = {10681--10692}, year = {2020}, editor = {III, Hal Daumé and Singh, Aarti}, volume = {119}, series = {Proceedings of Machine Learning Research}, month = {13--18 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v119/yang20b/yang20b.pdf}, url = {https://proceedings.mlr.press/v119/yang20b.html}, abstract = {Recently there has been growing interest in modeling sets with exchangeability such as point clouds. A shortcoming of current approaches is that they restrict the cardinality of the sets considered or can only express limited forms of distribution over unobserved data. To overcome these limitations, we introduce Energy-Based Processes (EBPs), which extend energy based models to exchangeable data while allowing neural network parameterizations of the energy function. A key advantage of these models is the ability to express more flexible distributions over sets without restricting their cardinality. We develop an efficient training procedure for EBPs that demonstrates state-of-the-art performance on a variety of tasks such as point cloud generation, classification, denoising, and image completion} }
Endnote
%0 Conference Paper %T Energy-Based Processes for Exchangeable Data %A Mengjiao Yang %A Bo Dai %A Hanjun Dai %A Dale Schuurmans %B Proceedings of the 37th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2020 %E Hal Daumé III %E Aarti Singh %F pmlr-v119-yang20b %I PMLR %P 10681--10692 %U https://proceedings.mlr.press/v119/yang20b.html %V 119 %X Recently there has been growing interest in modeling sets with exchangeability such as point clouds. A shortcoming of current approaches is that they restrict the cardinality of the sets considered or can only express limited forms of distribution over unobserved data. To overcome these limitations, we introduce Energy-Based Processes (EBPs), which extend energy based models to exchangeable data while allowing neural network parameterizations of the energy function. A key advantage of these models is the ability to express more flexible distributions over sets without restricting their cardinality. We develop an efficient training procedure for EBPs that demonstrates state-of-the-art performance on a variety of tasks such as point cloud generation, classification, denoising, and image completion
APA
Yang, M., Dai, B., Dai, H. & Schuurmans, D.. (2020). Energy-Based Processes for Exchangeable Data. Proceedings of the 37th International Conference on Machine Learning, in Proceedings of Machine Learning Research 119:10681-10692 Available from https://proceedings.mlr.press/v119/yang20b.html.

Related Material