Towards Automatically-Tuned Neural Networks

Hector Mendoza, Aaron Klein, Matthias Feurer, Jost Tobias Springenberg, Frank Hutter
Proceedings of the Workshop on Automatic Machine Learning, PMLR 64:58-65, 2016.

Abstract

Recent advances in AutoML have led to automated tools that can compete with machine learning experts on supervised learning tasks. However, current AutoML tools do not yet support modern neural networks effectively. In this work, we present a first version of Auto-Net, which provides automatically-tuned feed-forward neural networks without any human intervention. We report results on datasets from the recent AutoML challenge showing that ensembling Auto-Net with Auto-sklearn often performs better than either alone, and report the first results on winning a competition dataset against human experts with automatically-tuned neural networks.

Cite this Paper

BibTeX
@InProceedings{pmlr-v64-mendoza_towards_2016, title = {Towards Automatically-Tuned Neural Networks}, author = {Mendoza, Hector and Klein, Aaron and Feurer, Matthias and Springenberg, Jost Tobias and Hutter, Frank}, booktitle = {Proceedings of the Workshop on Automatic Machine Learning}, pages = {58--65}, year = {2016}, editor = {Hutter, Frank and Kotthoff, Lars and Vanschoren, Joaquin}, volume = {64}, series = {Proceedings of Machine Learning Research}, address = {New York, New York, USA}, month = {24 Jun}, publisher = {PMLR}, pdf = {http://proceedings.mlr.press/v64/mendoza_towards_2016.pdf}, url = {https://proceedings.mlr.press/v64/mendoza_towards_2016.html}, abstract = {Recent advances in AutoML have led to automated tools that can compete with machine learning experts on supervised learning tasks. However, current AutoML tools do not yet support modern neural networks effectively. In this work, we present a first version of Auto-Net, which provides automatically-tuned feed-forward neural networks without any human intervention. We report results on datasets from the recent AutoML challenge showing that ensembling Auto-Net with Auto-sklearn often performs better than either alone, and report the first results on winning a competition dataset against human experts with automatically-tuned neural networks.} }
Endnote
%0 Conference Paper %T Towards Automatically-Tuned Neural Networks %A Hector Mendoza %A Aaron Klein %A Matthias Feurer %A Jost Tobias Springenberg %A Frank Hutter %B Proceedings of the Workshop on Automatic Machine Learning %C Proceedings of Machine Learning Research %D 2016 %E Frank Hutter %E Lars Kotthoff %E Joaquin Vanschoren %F pmlr-v64-mendoza_towards_2016 %I PMLR %P 58--65 %U https://proceedings.mlr.press/v64/mendoza_towards_2016.html %V 64 %X Recent advances in AutoML have led to automated tools that can compete with machine learning experts on supervised learning tasks. However, current AutoML tools do not yet support modern neural networks effectively. In this work, we present a first version of Auto-Net, which provides automatically-tuned feed-forward neural networks without any human intervention. We report results on datasets from the recent AutoML challenge showing that ensembling Auto-Net with Auto-sklearn often performs better than either alone, and report the first results on winning a competition dataset against human experts with automatically-tuned neural networks.
RIS
TY - CPAPER TI - Towards Automatically-Tuned Neural Networks AU - Hector Mendoza AU - Aaron Klein AU - Matthias Feurer AU - Jost Tobias Springenberg AU - Frank Hutter BT - Proceedings of the Workshop on Automatic Machine Learning DA - 2016/12/04 ED - Frank Hutter ED - Lars Kotthoff ED - Joaquin Vanschoren ID - pmlr-v64-mendoza_towards_2016 PB - PMLR DP - Proceedings of Machine Learning Research VL - 64 SP - 58 EP - 65 L1 - http://proceedings.mlr.press/v64/mendoza_towards_2016.pdf UR - https://proceedings.mlr.press/v64/mendoza_towards_2016.html AB - Recent advances in AutoML have led to automated tools that can compete with machine learning experts on supervised learning tasks. However, current AutoML tools do not yet support modern neural networks effectively. In this work, we present a first version of Auto-Net, which provides automatically-tuned feed-forward neural networks without any human intervention. We report results on datasets from the recent AutoML challenge showing that ensembling Auto-Net with Auto-sklearn often performs better than either alone, and report the first results on winning a competition dataset against human experts with automatically-tuned neural networks. ER -
APA
Mendoza, H., Klein, A., Feurer, M., Springenberg, J.T. & Hutter, F.. (2016). Towards Automatically-Tuned Neural Networks. Proceedings of the Workshop on Automatic Machine Learning, in Proceedings of Machine Learning Research 64:58-65 Available from https://proceedings.mlr.press/v64/mendoza_towards_2016.html.

Related Material