Measuring the Data Efficiency of Deep Learning Methods

Hlynur Hlynsson, Alberto Escalante-B., Laurenz Wiskott

2019

Abstract

In this paper, we propose a new experimental protocol and use it to benchmark the data efficiency — performance as a function of training set size — of two deep learning algorithms, convolutional neural networks (CNNs) and hierarchical information-preserving graph-based slow feature analysis (HiGSFA), for tasks in classification and transfer learning scenarios. The algorithms are trained on different-sized subsets of the MNIST and Omniglot data sets. HiGSFA outperforms standard CNN networks when the models are trained on 50 and 200 samples per class for MNIST classification. In other cases, the CNNs perform better. The results suggest that there are cases where greedy, locally optimal bottom-up learning is equally or more powerful than global gradient-based learning.

Paper Citation

in Harvard Style

Hlynsson H., Escalante-B. A. and Wiskott L. (2019). Measuring the Data Efficiency of Deep Learning Methods.In Proceedings of the 8th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM, ISBN 978-989-758-351-3, pages 691-698. DOI: 10.5220/0007456306910698

in Bibtex Style

@conference{icpram19,
author={Hlynur Hlynsson and Alberto Escalante-B. and Laurenz Wiskott},
title={Measuring the Data Efficiency of Deep Learning Methods},
booktitle={Proceedings of the 8th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM,},
year={2019},
pages={691-698},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0007456306910698},
isbn={978-989-758-351-3},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 8th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM,
TI - Measuring the Data Efficiency of Deep Learning Methods
SN - 978-989-758-351-3
AU - Hlynsson H.
AU - Escalante-B. A.
AU - Wiskott L.
PY - 2019
SP - 691
EP - 698
DO - 10.5220/0007456306910698