MLtuner automatically tunes settings for training tunables (such as the
learning rate, the momentum, the mini-batch size, and the data staleness bound)
that have a significant impact on large-scale machine learning (ML)
performance. Traditionally, these tunables are set manually, which is
unsurprisingly error-prone and difficult to do without extensive domain
knowledge. MLtuner uses efficient snapshotting, branching, and
optimization-guided online trial-and-error to find good initial settings as
well as to re-tune settings during execution. Experiments show that MLtuner can
robustly find and re-tune tunable settings for a variety of ML applications,
including image classification (for 3 models and 2 datasets), video
classification, and matrix factorization. Compared to state-of-the-art ML
auto-tuning approaches, MLtuner is more robust for large problems and over an
order of magnitude faster.
Description
[1803.07445] MLtuner: System Support for Automatic Machine Learning Tuning
%0 Generic
%1 cui2018mltuner
%A Cui, Henggang
%A Ganger, Gregory R.
%A Gibbons, Phillip B.
%D 2018
%K hyperparameter tuning
%T MLtuner: System Support for Automatic Machine Learning Tuning
%U http://arxiv.org/abs/1803.07445
%X MLtuner automatically tunes settings for training tunables (such as the
learning rate, the momentum, the mini-batch size, and the data staleness bound)
that have a significant impact on large-scale machine learning (ML)
performance. Traditionally, these tunables are set manually, which is
unsurprisingly error-prone and difficult to do without extensive domain
knowledge. MLtuner uses efficient snapshotting, branching, and
optimization-guided online trial-and-error to find good initial settings as
well as to re-tune settings during execution. Experiments show that MLtuner can
robustly find and re-tune tunable settings for a variety of ML applications,
including image classification (for 3 models and 2 datasets), video
classification, and matrix factorization. Compared to state-of-the-art ML
auto-tuning approaches, MLtuner is more robust for large problems and over an
order of magnitude faster.
@misc{cui2018mltuner,
abstract = {MLtuner automatically tunes settings for training tunables (such as the
learning rate, the momentum, the mini-batch size, and the data staleness bound)
that have a significant impact on large-scale machine learning (ML)
performance. Traditionally, these tunables are set manually, which is
unsurprisingly error-prone and difficult to do without extensive domain
knowledge. MLtuner uses efficient snapshotting, branching, and
optimization-guided online trial-and-error to find good initial settings as
well as to re-tune settings during execution. Experiments show that MLtuner can
robustly find and re-tune tunable settings for a variety of ML applications,
including image classification (for 3 models and 2 datasets), video
classification, and matrix factorization. Compared to state-of-the-art ML
auto-tuning approaches, MLtuner is more robust for large problems and over an
order of magnitude faster.},
added-at = {2018-03-21T15:50:34.000+0100},
author = {Cui, Henggang and Ganger, Gregory R. and Gibbons, Phillip B.},
biburl = {https://www.bibsonomy.org/bibtex/20ff64e8de6f9145a2f125ff9b67b575b/rcb},
description = {[1803.07445] MLtuner: System Support for Automatic Machine Learning Tuning},
interhash = {6fbe73314c16656bf9fc33c73e8a7591},
intrahash = {0ff64e8de6f9145a2f125ff9b67b575b},
keywords = {hyperparameter tuning},
note = {cite arxiv:1803.07445},
timestamp = {2018-03-21T15:50:34.000+0100},
title = {MLtuner: System Support for Automatic Machine Learning Tuning},
url = {http://arxiv.org/abs/1803.07445},
year = 2018
}