Learning and Selecting Features Jointly with Point-wise Gated Boltzmann Machines

Kihyuk Sohn, Guanyu Zhou, Chansoo Lee, Honglak Lee
Proceedings of the 30th International Conference on Machine Learning, PMLR 28(2):217-225, 2013.

Abstract

Unsupervised feature learning has emerged as a promising tool in learning representations from unlabeled data. However, it is still challenging to learn useful high-level features when the data contains a significant amount of irrelevant patterns. Although feature selection can be used for such complex data, it may fail when we have to build a learning system from scratch (i.e., starting from the lack of useful raw features). To address this problem, we propose a point-wise gated Boltzmann machine, a unified generative model that combines feature learning and feature selection. Our model performs not only feature selection on learned high-level features (i.e., hidden units), but also dynamic feature selection on raw features (i.e., visible units) through a gating mechanism. For each example, the model can adaptively focus on a variable subset of visible nodes corresponding to the task-relevant patterns, while ignoring the visible units corresponding to the task-irrelevant patterns. In experiments, our method achieves improved performance over state-of-the-art in several visual recognition benchmarks.

Cite this Paper


BibTeX
@InProceedings{pmlr-v28-sohn13, title = {Learning and Selecting Features Jointly with Point-wise Gated {B}oltzmann Machines}, author = {Sohn, Kihyuk and Zhou, Guanyu and Lee, Chansoo and Lee, Honglak}, booktitle = {Proceedings of the 30th International Conference on Machine Learning}, pages = {217--225}, year = {2013}, editor = {Dasgupta, Sanjoy and McAllester, David}, volume = {28}, number = {2}, series = {Proceedings of Machine Learning Research}, address = {Atlanta, Georgia, USA}, month = {17--19 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v28/sohn13.pdf}, url = {https://proceedings.mlr.press/v28/sohn13.html}, abstract = {Unsupervised feature learning has emerged as a promising tool in learning representations from unlabeled data. However, it is still challenging to learn useful high-level features when the data contains a significant amount of irrelevant patterns. Although feature selection can be used for such complex data, it may fail when we have to build a learning system from scratch (i.e., starting from the lack of useful raw features). To address this problem, we propose a point-wise gated Boltzmann machine, a unified generative model that combines feature learning and feature selection. Our model performs not only feature selection on learned high-level features (i.e., hidden units), but also dynamic feature selection on raw features (i.e., visible units) through a gating mechanism. For each example, the model can adaptively focus on a variable subset of visible nodes corresponding to the task-relevant patterns, while ignoring the visible units corresponding to the task-irrelevant patterns. In experiments, our method achieves improved performance over state-of-the-art in several visual recognition benchmarks. } }
Endnote
%0 Conference Paper %T Learning and Selecting Features Jointly with Point-wise Gated Boltzmann Machines %A Kihyuk Sohn %A Guanyu Zhou %A Chansoo Lee %A Honglak Lee %B Proceedings of the 30th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2013 %E Sanjoy Dasgupta %E David McAllester %F pmlr-v28-sohn13 %I PMLR %P 217--225 %U https://proceedings.mlr.press/v28/sohn13.html %V 28 %N 2 %X Unsupervised feature learning has emerged as a promising tool in learning representations from unlabeled data. However, it is still challenging to learn useful high-level features when the data contains a significant amount of irrelevant patterns. Although feature selection can be used for such complex data, it may fail when we have to build a learning system from scratch (i.e., starting from the lack of useful raw features). To address this problem, we propose a point-wise gated Boltzmann machine, a unified generative model that combines feature learning and feature selection. Our model performs not only feature selection on learned high-level features (i.e., hidden units), but also dynamic feature selection on raw features (i.e., visible units) through a gating mechanism. For each example, the model can adaptively focus on a variable subset of visible nodes corresponding to the task-relevant patterns, while ignoring the visible units corresponding to the task-irrelevant patterns. In experiments, our method achieves improved performance over state-of-the-art in several visual recognition benchmarks.
RIS
TY - CPAPER TI - Learning and Selecting Features Jointly with Point-wise Gated Boltzmann Machines AU - Kihyuk Sohn AU - Guanyu Zhou AU - Chansoo Lee AU - Honglak Lee BT - Proceedings of the 30th International Conference on Machine Learning DA - 2013/05/13 ED - Sanjoy Dasgupta ED - David McAllester ID - pmlr-v28-sohn13 PB - PMLR DP - Proceedings of Machine Learning Research VL - 28 IS - 2 SP - 217 EP - 225 L1 - http://proceedings.mlr.press/v28/sohn13.pdf UR - https://proceedings.mlr.press/v28/sohn13.html AB - Unsupervised feature learning has emerged as a promising tool in learning representations from unlabeled data. However, it is still challenging to learn useful high-level features when the data contains a significant amount of irrelevant patterns. Although feature selection can be used for such complex data, it may fail when we have to build a learning system from scratch (i.e., starting from the lack of useful raw features). To address this problem, we propose a point-wise gated Boltzmann machine, a unified generative model that combines feature learning and feature selection. Our model performs not only feature selection on learned high-level features (i.e., hidden units), but also dynamic feature selection on raw features (i.e., visible units) through a gating mechanism. For each example, the model can adaptively focus on a variable subset of visible nodes corresponding to the task-relevant patterns, while ignoring the visible units corresponding to the task-irrelevant patterns. In experiments, our method achieves improved performance over state-of-the-art in several visual recognition benchmarks. ER -
APA
Sohn, K., Zhou, G., Lee, C. & Lee, H.. (2013). Learning and Selecting Features Jointly with Point-wise Gated Boltzmann Machines. Proceedings of the 30th International Conference on Machine Learning, in Proceedings of Machine Learning Research 28(2):217-225 Available from https://proceedings.mlr.press/v28/sohn13.html.

Related Material