Learning Low-Dimensional Temporal Representations

Bing Su, Ying Wu
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:4761-4770, 2018.

Abstract

Low-dimensional discriminative representations enhance machine learning methods in both performance and complexity, motivating supervised dimensionality reduction (DR) that transforms high-dimensional data to a discriminative subspace. Most DR methods require data to be i.i.d., however, in some domains, data naturally come in sequences, where the observations are temporally correlated. We propose a DR method called LT-LDA to learn low-dimensional temporal representations. We construct the separability among sequence classes by lifting the holistic temporal structures, which are established based on temporal alignments and may change in different subspaces. We jointly learn the subspace and the associated alignments by optimizing an objective which favors easily-separable temporal structures, and show that this objective is connected to the inference of alignments, thus allows an iterative solution. We provide both theoretical insight and empirical evaluation on real-world sequence datasets to show the interest of our method.

Cite this Paper

BibTeX
@InProceedings{pmlr-v80-su18a, title = {Learning Low-Dimensional Temporal Representations}, author = {Su, Bing and Wu, Ying}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {4761--4770}, year = {2018}, editor = {Dy, Jennifer and Krause, Andreas}, volume = {80}, series = {Proceedings of Machine Learning Research}, month = {10--15 Jul}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v80/su18a/su18a.pdf}, url = {https://proceedings.mlr.press/v80/su18a.html}, abstract = {Low-dimensional discriminative representations enhance machine learning methods in both performance and complexity, motivating supervised dimensionality reduction (DR) that transforms high-dimensional data to a discriminative subspace. Most DR methods require data to be i.i.d., however, in some domains, data naturally come in sequences, where the observations are temporally correlated. We propose a DR method called LT-LDA to learn low-dimensional temporal representations. We construct the separability among sequence classes by lifting the holistic temporal structures, which are established based on temporal alignments and may change in different subspaces. We jointly learn the subspace and the associated alignments by optimizing an objective which favors easily-separable temporal structures, and show that this objective is connected to the inference of alignments, thus allows an iterative solution. We provide both theoretical insight and empirical evaluation on real-world sequence datasets to show the interest of our method.} }
Endnote
%0 Conference Paper %T Learning Low-Dimensional Temporal Representations %A Bing Su %A Ying Wu %B Proceedings of the 35th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2018 %E Jennifer Dy %E Andreas Krause %F pmlr-v80-su18a %I PMLR %P 4761--4770 %U https://proceedings.mlr.press/v80/su18a.html %V 80 %X Low-dimensional discriminative representations enhance machine learning methods in both performance and complexity, motivating supervised dimensionality reduction (DR) that transforms high-dimensional data to a discriminative subspace. Most DR methods require data to be i.i.d., however, in some domains, data naturally come in sequences, where the observations are temporally correlated. We propose a DR method called LT-LDA to learn low-dimensional temporal representations. We construct the separability among sequence classes by lifting the holistic temporal structures, which are established based on temporal alignments and may change in different subspaces. We jointly learn the subspace and the associated alignments by optimizing an objective which favors easily-separable temporal structures, and show that this objective is connected to the inference of alignments, thus allows an iterative solution. We provide both theoretical insight and empirical evaluation on real-world sequence datasets to show the interest of our method.
APA
Su, B. & Wu, Y.. (2018). Learning Low-Dimensional Temporal Representations. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:4761-4770 Available from https://proceedings.mlr.press/v80/su18a.html.

Related Material