EuclidNets: Combining Hardware and Architecture Design for Efficient Training and Inference

Mariana Prazeres, Mariana Prazeres, Xinlin Li, Adam Oberman, Vahid Nia

2022

Abstract

In order to deploy deep neural networks on edge devices, compressed (resource efficient) networks need to be developed. While established compression methods, such as quantization, pruning, and architecture search are designed for conventional hardware, further gains are possible if compressed architectures are coupled with novel hardware designs. In this work, we propose EuclidNet, a compressed network designed to be implemented on hardware which replaces multiplication, wx, with squared difference (x − w)2. EuclidNet allows for a low precision hardware implementation which is about twice as efficient (in term of logic gate counts) as the comparable conventional hardware, with acceptably small loss of accuracy. Moreover, the network can be trained and quantized using standard methods, without requiring additional training time. Codes and pre-trained models are available.

Paper Citation

in Harvard Style

Prazeres M., Li X., Oberman A. and Nia V. (2022). EuclidNets: Combining Hardware and Architecture Design for Efficient Training and Inference. In Proceedings of the 11th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM, ISBN 978-989-758-549-4, pages 141-151. DOI: 10.5220/0010988500003122

in Bibtex Style

@conference{icpram22,
author={Mariana Prazeres and Xinlin Li and Adam Oberman and Vahid Nia},
title={EuclidNets: Combining Hardware and Architecture Design for Efficient Training and Inference},
booktitle={Proceedings of the 11th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM,},
year={2022},
pages={141-151},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0010988500003122},
isbn={978-989-758-549-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 11th International Conference on Pattern Recognition Applications and Methods - Volume 1: ICPRAM,
TI - EuclidNets: Combining Hardware and Architecture Design for Efficient Training and Inference
SN - 978-989-758-549-4
AU - Prazeres M.
AU - Li X.
AU - Oberman A.
AU - Nia V.
PY - 2022
SP - 141
EP - 151
DO - 10.5220/0010988500003122