Standard

Applicability of Minifloats for Efficient Calculations in Neural Networks. / Kondrat’ev, A. Yu; Goncharenko, A. I.

In: Optoelectronics, Instrumentation and Data Processing, Vol. 56, No. 1, 01.01.2020, p. 76-80.

Research output: Contribution to journal › Article › peer-review

Harvard

Kondrat’ev, AY & Goncharenko, AI 2020, 'Applicability of Minifloats for Efficient Calculations in Neural Networks', Optoelectronics, Instrumentation and Data Processing, vol. 56, no. 1, pp. 76-80. https://doi.org/10.3103/S8756699020010100

APA

Kondrat’ev, A. Y., & Goncharenko, A. I. (2020). Applicability of Minifloats for Efficient Calculations in Neural Networks. Optoelectronics, Instrumentation and Data Processing, 56(1), 76-80. https://doi.org/10.3103/S8756699020010100

Vancouver

Kondrat’ev AY, Goncharenko AI. Applicability of Minifloats for Efficient Calculations in Neural Networks. Optoelectronics, Instrumentation and Data Processing. 2020 Jan 1;56(1):76-80. doi: 10.3103/S8756699020010100

Author

Kondrat’ev, A. Yu ; Goncharenko, A. I. / Applicability of Minifloats for Efficient Calculations in Neural Networks. In: Optoelectronics, Instrumentation and Data Processing. 2020 ; Vol. 56, No. 1. pp. 76-80.

BibTeX

@article{dcdb2ef7fc8040fbae1af024f31df31f,
title = "Applicability of Minifloats for Efficient Calculations in Neural Networks",
abstract = "The possibility of the inference of neural networks on minifloats has been studied. Calculations using a float16 accumulator for intermediate computing were performed. Performance was tested on the GoogleNet, ResNet-50, and MobileNet-v2 convolutional neural network and the DeepSpeechv01 recurrent network. The experiments showed that the performance of these neural networks with 11-bit minifloats is not inferior to the performance of networks with the float32 standard type without additional training. The results indicate that minifloats can be used to design efficient computers for the inference of neural networks.",
keywords = "data types, deep learning, minifloat, neural networks, special-purpose computers",
author = "Kondrat{\textquoteright}ev, {A. Yu} and Goncharenko, {A. I.}",
year = "2020",
month = jan,
day = "1",
doi = "10.3103/S8756699020010100",
language = "English",
volume = "56",
pages = "76--80",
journal = "Optoelectronics, Instrumentation and Data Processing",
issn = "8756-6990",
publisher = "Allerton Press Inc.",
number = "1",

}

RIS

TY - JOUR

T1 - Applicability of Minifloats for Efficient Calculations in Neural Networks

AU - Kondrat’ev, A. Yu

AU - Goncharenko, A. I.

PY - 2020/1/1

Y1 - 2020/1/1

N2 - The possibility of the inference of neural networks on minifloats has been studied. Calculations using a float16 accumulator for intermediate computing were performed. Performance was tested on the GoogleNet, ResNet-50, and MobileNet-v2 convolutional neural network and the DeepSpeechv01 recurrent network. The experiments showed that the performance of these neural networks with 11-bit minifloats is not inferior to the performance of networks with the float32 standard type without additional training. The results indicate that minifloats can be used to design efficient computers for the inference of neural networks.

AB - The possibility of the inference of neural networks on minifloats has been studied. Calculations using a float16 accumulator for intermediate computing were performed. Performance was tested on the GoogleNet, ResNet-50, and MobileNet-v2 convolutional neural network and the DeepSpeechv01 recurrent network. The experiments showed that the performance of these neural networks with 11-bit minifloats is not inferior to the performance of networks with the float32 standard type without additional training. The results indicate that minifloats can be used to design efficient computers for the inference of neural networks.

KW - data types

KW - deep learning

KW - minifloat

KW - neural networks

KW - special-purpose computers

UR - http://www.scopus.com/inward/record.url?scp=85089108507&partnerID=8YFLogxK

U2 - 10.3103/S8756699020010100

DO - 10.3103/S8756699020010100

M3 - Article

AN - SCOPUS:85089108507

VL - 56

SP - 76

EP - 80

JO - Optoelectronics, Instrumentation and Data Processing

JF - Optoelectronics, Instrumentation and Data Processing

SN - 8756-6990

IS - 1

ER -

ID: 24961989