Training Neural Networks in Single vs. Double Precision

Tomas Hrycej, Bernhard Bermeitinger, Siegfried Handschuh

2022

Abstract

The commitment to single-precision floating-point arithmetic is widespread in the deep learning community. To evaluate whether this commitment is justified, the influence of computing precision (single and double precision) on the optimization performance of the Conjugate Gradient (CG) method (a second-order optimization algorithm) and Root Mean Square Propagation (RMSprop) (a first-order algorithm) has been investigated. Tests of neural networks with one to five fully connected hidden layers and moderate or strong nonlinearity with up to 4 million network parameters have been optimized for Mean Square Error (MSE). The training tasks have been set up so that their MSE minimum was known to be zero. Computing experiments have dis-closed that single-precision can keep up (with superlinear convergence) with double-precision as long as line search finds an improvement. First-order methods such as RMSprop do not benefit from double precision. However, for moderately nonlinear tasks, CG is clearly superior. For strongly nonlinear tasks, both algorithm classes find only solutions fairly poor in terms of mean square error as related to the output variance. CG with double floating-point precision is superior whenever the solutions have the potential to be useful for the application goal.

Paper Citation

in Harvard Style

Hrycej T., Bermeitinger B. and Handschuh S. (2022). Training Neural Networks in Single vs. Double Precision. In Proceedings of the 14th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2022) - Volume 1: KDIR; ISBN 978-989-758-614-9, SciTePress, pages 307-314. DOI: 10.5220/0011577900003335

in Bibtex Style

@conference{kdir22,
author={Tomas Hrycej and Bernhard Bermeitinger and Siegfried Handschuh},
title={Training Neural Networks in Single vs. Double Precision},
booktitle={Proceedings of the 14th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2022) - Volume 1: KDIR},
year={2022},
pages={307-314},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0011577900003335},
isbn={978-989-758-614-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 14th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2022) - Volume 1: KDIR
TI - Training Neural Networks in Single vs. Double Precision
SN - 978-989-758-614-9
AU - Hrycej T.
AU - Bermeitinger B.
AU - Handschuh S.
PY - 2022
SP - 307
EP - 314
DO - 10.5220/0011577900003335
PB - SciTePress