Deep Factors for Forecasting

Yuyang Wang, Alex Smola, Danielle Maddix, Jan Gasthaus, Dean Foster, Tim Januschowski
Proceedings of the 36th International Conference on Machine Learning, PMLR 97:6607-6617, 2019.

Abstract

Producing probabilistic forecasts for large collections of similar and/or dependent time series is a practically highly relevant, yet challenging task. Classical time series models fail to capture complex patterns in the data and multivariate techniques struggle to scale to large problem sizes, but their reliance on strong structural assumptions makes them data-efficient and allows them to provide estimates of uncertainty. The converse is true for models based on deep neural networks, which can learn complex patterns and dependencies given enough data. In this paper, we propose a hybrid model that incorporates the benefits of both approaches. Our new method is data-driven and scalable via a latent, global, deep component. It also handles uncertainty through a local classical model. We provide both theoretical and empirical evidence for the soundness of our approach through a necessary and sufficient decomposition of exchangeable time series into a global and a local part and extensive experiments. Our experiments demonstrate the advantages of our model both in term of data efficiency and computational complexity.

Cite this Paper


BibTeX
@InProceedings{pmlr-v97-wang19k, title = {Deep Factors for Forecasting}, author = {Wang, Yuyang and Smola, Alex and Maddix, Danielle and Gasthaus, Jan and Foster, Dean and Januschowski, Tim}, booktitle = {Proceedings of the 36th International Conference on Machine Learning}, pages = {6607--6617}, year = {2019}, editor = {Chaudhuri, Kamalika and Salakhutdinov, Ruslan}, volume = {97}, series = {Proceedings of Machine Learning Research}, month = {09--15 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v97/wang19k/wang19k.pdf}, url = {https://proceedings.mlr.press/v97/wang19k.html}, abstract = {Producing probabilistic forecasts for large collections of similar and/or dependent time series is a practically highly relevant, yet challenging task. Classical time series models fail to capture complex patterns in the data and multivariate techniques struggle to scale to large problem sizes, but their reliance on strong structural assumptions makes them data-efficient and allows them to provide estimates of uncertainty. The converse is true for models based on deep neural networks, which can learn complex patterns and dependencies given enough data. In this paper, we propose a hybrid model that incorporates the benefits of both approaches. Our new method is data-driven and scalable via a latent, global, deep component. It also handles uncertainty through a local classical model. We provide both theoretical and empirical evidence for the soundness of our approach through a necessary and sufficient decomposition of exchangeable time series into a global and a local part and extensive experiments. Our experiments demonstrate the advantages of our model both in term of data efficiency and computational complexity.} }
Endnote
%0 Conference Paper %T Deep Factors for Forecasting %A Yuyang Wang %A Alex Smola %A Danielle Maddix %A Jan Gasthaus %A Dean Foster %A Tim Januschowski %B Proceedings of the 36th International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2019 %E Kamalika Chaudhuri %E Ruslan Salakhutdinov %F pmlr-v97-wang19k %I PMLR %P 6607--6617 %U https://proceedings.mlr.press/v97/wang19k.html %V 97 %X Producing probabilistic forecasts for large collections of similar and/or dependent time series is a practically highly relevant, yet challenging task. Classical time series models fail to capture complex patterns in the data and multivariate techniques struggle to scale to large problem sizes, but their reliance on strong structural assumptions makes them data-efficient and allows them to provide estimates of uncertainty. The converse is true for models based on deep neural networks, which can learn complex patterns and dependencies given enough data. In this paper, we propose a hybrid model that incorporates the benefits of both approaches. Our new method is data-driven and scalable via a latent, global, deep component. It also handles uncertainty through a local classical model. We provide both theoretical and empirical evidence for the soundness of our approach through a necessary and sufficient decomposition of exchangeable time series into a global and a local part and extensive experiments. Our experiments demonstrate the advantages of our model both in term of data efficiency and computational complexity.
APA
Wang, Y., Smola, A., Maddix, D., Gasthaus, J., Foster, D. & Januschowski, T.. (2019). Deep Factors for Forecasting. Proceedings of the 36th International Conference on Machine Learning, in Proceedings of Machine Learning Research 97:6607-6617 Available from https://proceedings.mlr.press/v97/wang19k.html.

Related Material