Compression of Activation Signals from Split Deep Neural Network
(2022) 14th IEEE Latin-American Conference on Communications, LATINCOM 2022 In 2022 IEEE Latin-American Conference on Communications, LATINCOM 2022- Abstract
The use of artificial neural networks for the purpose of image classification, together with the advancement in computational capabilities of edge devices, plays an important role in the new emerging 5G use case scenarios. However, one of the main challenges of applications involving the use of these networks in edge devices is still the limitation of computational resources. An alternative for saving resources and promoting privacy are the split learning (or split inference) techniques, in which a deep neural network is cut into two parts and executed in distinct devices. Most of these techniques rely on sending the activation signals (output of the cut layer) through the communication channel. This work proposes a new compression... (More)
The use of artificial neural networks for the purpose of image classification, together with the advancement in computational capabilities of edge devices, plays an important role in the new emerging 5G use case scenarios. However, one of the main challenges of applications involving the use of these networks in edge devices is still the limitation of computational resources. An alternative for saving resources and promoting privacy are the split learning (or split inference) techniques, in which a deep neural network is cut into two parts and executed in distinct devices. Most of these techniques rely on sending the activation signals (output of the cut layer) through the communication channel. This work proposes a new compression algorithm for decreasing the bit rate required for the transmission of the activation signals (or simply 'scores'). The presented results demonstrate that the transmission rate can be decreased without hurting the neural network accuracy.
(Less)
- author
- Brito, Flavio LU ; Silva, Lucas ; Ramalho, Leonardo ; Lins, Silvia ; Linder, Neiva and Klautau, Aldebaro
- organization
- publishing date
- 2022
- type
- Chapter in Book/Report/Conference proceeding
- publication status
- published
- subject
- keywords
- deep neural networks, image classification, scores compression, Split learning
- host publication
- 2022 IEEE Latin-American Conference on Communications, LATINCOM 2022
- series title
- 2022 IEEE Latin-American Conference on Communications, LATINCOM 2022
- editor
- Moraes, Igor M. ; Campista, Miguel Elias M. ; Ghamri-Doudane, Yacine ; Luis Henrique M. K. Costa, Costa and Rubinstein, Marcelo G.
- publisher
- IEEE - Institute of Electrical and Electronics Engineers Inc.
- conference name
- 14th IEEE Latin-American Conference on Communications, LATINCOM 2022
- conference location
- Rio de Janeiro, Brazil
- conference dates
- 2022-11-30 - 2022-12-02
- external identifiers
-
- scopus:85146685598
- ISBN
- 9781665482257
- DOI
- 10.1109/LATINCOM56090.2022.10000526
- language
- English
- LU publication?
- yes
- id
- 64c5c8d0-749c-430e-9ecd-02c3b48b2039
- date added to LUP
- 2023-02-15 15:05:36
- date last changed
- 2023-02-15 15:05:36
@inproceedings{64c5c8d0-749c-430e-9ecd-02c3b48b2039,
abstract = {{<p>The use of artificial neural networks for the purpose of image classification, together with the advancement in computational capabilities of edge devices, plays an important role in the new emerging 5G use case scenarios. However, one of the main challenges of applications involving the use of these networks in edge devices is still the limitation of computational resources. An alternative for saving resources and promoting privacy are the split learning (or split inference) techniques, in which a deep neural network is cut into two parts and executed in distinct devices. Most of these techniques rely on sending the activation signals (output of the cut layer) through the communication channel. This work proposes a new compression algorithm for decreasing the bit rate required for the transmission of the activation signals (or simply 'scores'). The presented results demonstrate that the transmission rate can be decreased without hurting the neural network accuracy.</p>}},
author = {{Brito, Flavio and Silva, Lucas and Ramalho, Leonardo and Lins, Silvia and Linder, Neiva and Klautau, Aldebaro}},
booktitle = {{2022 IEEE Latin-American Conference on Communications, LATINCOM 2022}},
editor = {{Moraes, Igor M. and Campista, Miguel Elias M. and Ghamri-Doudane, Yacine and Luis Henrique M. K. Costa, Costa and Rubinstein, Marcelo G.}},
isbn = {{9781665482257}},
keywords = {{deep neural networks; image classification; scores compression; Split learning}},
language = {{eng}},
publisher = {{IEEE - Institute of Electrical and Electronics Engineers Inc.}},
series = {{2022 IEEE Latin-American Conference on Communications, LATINCOM 2022}},
title = {{Compression of Activation Signals from Split Deep Neural Network}},
url = {{http://dx.doi.org/10.1109/LATINCOM56090.2022.10000526}},
doi = {{10.1109/LATINCOM56090.2022.10000526}},
year = {{2022}},
}