Transformation Autoregressive Networks

Junier Oliva, Avinava Dubey, Manzil Zaheer, Barnabas Poczos, Ruslan Salakhutdinov, Eric Xing, Jeff Schneider
Proceedings of the 35th International Conference on Machine Learning, PMLR 80:3898-3907, 2018.

Abstract

The fundamental task of general density estimation $p(x)$ has been of keen interest to machine learning. In this work, we attempt to systematically characterize methods for density estimation. Broadly speaking, most of the existing methods can be categorized into either using: a) autoregressive models to estimate the conditional factors of the chain rule, $p(x_{i}\, |\, x_{i-1}, \ldots)$; or b) non-linear transformations of variables of a simple base distribution. Based on the study of the characteristics of these categories, we propose multiple novel methods for each category. For example we propose RNN based transformations to model non-Markovian dependencies. Further, through a comprehensive study over both real world and synthetic data, we show that jointly leveraging transformations of variables and autoregressive conditional models, results in a considerable improvement in performance. We illustrate the use of our models in outlier detection and image modeling. Finally we introduce a novel data driven framework for learning a family of distributions.

Cite this Paper

BibTeX
@InProceedings{pmlr-v80-oliva18a, title = {Transformation Autoregressive Networks}, author = {Oliva, Junier and Dubey, Avinava and Zaheer, Manzil and Poczos, Barnabas and Salakhutdinov, Ruslan and Xing, Eric and Schneider, Jeff}, booktitle = {Proceedings of the 35th International Conference on Machine Learning}, pages = {3898--3907}, 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/oliva18a/oliva18a.pdf}, url = {https://proceedings.mlr.press/v80/oliva18a.html}, abstract = {The fundamental task of general density estimation $p(x)$ has been of keen interest to machine learning. In this work, we attempt to systematically characterize methods for density estimation. Broadly speaking, most of the existing methods can be categorized into either using: a) autoregressive models to estimate the conditional factors of the chain rule, $p(x_{i}\, |\, x_{i-1}, \ldots)$; or b) non-linear transformations of variables of a simple base distribution. Based on the study of the characteristics of these categories, we propose multiple novel methods for each category. For example we propose RNN based transformations to model non-Markovian dependencies. Further, through a comprehensive study over both real world and synthetic data, we show that jointly leveraging transformations of variables and autoregressive conditional models, results in a considerable improvement in performance. We illustrate the use of our models in outlier detection and image modeling. Finally we introduce a novel data driven framework for learning a family of distributions.} }
Endnote
%0 Conference Paper %T Transformation Autoregressive Networks %A Junier Oliva %A Avinava Dubey %A Manzil Zaheer %A Barnabas Poczos %A Ruslan Salakhutdinov %A Eric Xing %A Jeff Schneider %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-oliva18a %I PMLR %P 3898--3907 %U https://proceedings.mlr.press/v80/oliva18a.html %V 80 %X The fundamental task of general density estimation $p(x)$ has been of keen interest to machine learning. In this work, we attempt to systematically characterize methods for density estimation. Broadly speaking, most of the existing methods can be categorized into either using: a) autoregressive models to estimate the conditional factors of the chain rule, $p(x_{i}\, |\, x_{i-1}, \ldots)$; or b) non-linear transformations of variables of a simple base distribution. Based on the study of the characteristics of these categories, we propose multiple novel methods for each category. For example we propose RNN based transformations to model non-Markovian dependencies. Further, through a comprehensive study over both real world and synthetic data, we show that jointly leveraging transformations of variables and autoregressive conditional models, results in a considerable improvement in performance. We illustrate the use of our models in outlier detection and image modeling. Finally we introduce a novel data driven framework for learning a family of distributions.
APA
Oliva, J., Dubey, A., Zaheer, M., Poczos, B., Salakhutdinov, R., Xing, E. & Schneider, J.. (2018). Transformation Autoregressive Networks. Proceedings of the 35th International Conference on Machine Learning, in Proceedings of Machine Learning Research 80:3898-3907 Available from https://proceedings.mlr.press/v80/oliva18a.html.

Related Material