Work-in-progress: BPNet: Branch-pruned conditional neural network for systematic time-accuracy tradeoff in DNN inference

Kyungchul Park; Youngmin Yi

doi:10.1145/3349567.3351721

Work-in-progress: BPNet: Branch-pruned conditional neural network for systematic time-accuracy tradeoff in DNN inference

Kyungchul Park
, Youngmin Yi

School of Electrical and Computer Engineering

University of Seoul

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

2 Scopus citations

Abstract

Recently, there have been attempts to execute the neural network conditionally with auxiliary classifiers allowing early termination depending on the difficulty of the input, which can reduce the execution time or energy consumption without any or with negligible accuracy decrease. However, these studies do not consider how many or where the auxiliary classifiers, or branches, should be added in a systematic fashion. In this paper, we propose Branch-pruned Conditional Neural Network (BPNet) and its methodology in which the time-accuracy tradeoff for the conditional neural network can be found systematically. We applied BPNet to SqueezeNet, ResNet-20, and VGG-16 with CIFAR-10 and 100. BPNet achieves on average 2.0× of speedups without any accuracy drop on average compared to the base network.

Original language	English
Title of host publication	Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019
Publisher	Association for Computing Machinery, Inc
ISBN (Electronic)	9781450369237
DOIs	https://doi.org/10.1145/3349567.3351721
State	Published - 13 Oct 2019
Event	2019 International Conference on Hardware/Software Codesign and System Synthesis, CODES/ISSS 2019 - New York, United States Duration: 13 Oct 2019 → 18 Oct 2019

Publication series

Name	Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019

Conference

Conference	2019 International Conference on Hardware/Software Codesign and System Synthesis, CODES/ISSS 2019
Country/Territory	United States
City	New York
Period	13/10/19 → 18/10/19

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 7 Affordable and Clean Energy

Access to Document

10.1145/3349567.3351721

Cite this

Park, K., & Yi, Y. (2019). Work-in-progress: BPNet: Branch-pruned conditional neural network for systematic time-accuracy tradeoff in DNN inference. In Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019 Article a2 (Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019). Association for Computing Machinery, Inc. https://doi.org/10.1145/3349567.3351721

Park, Kyungchul ; Yi, Youngmin. / Work-in-progress : BPNet: Branch-pruned conditional neural network for systematic time-accuracy tradeoff in DNN inference. Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019. Association for Computing Machinery, Inc, 2019. (Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019).

@inproceedings{d28417ee6b5e4756b48df29e452c924f,

title = "Work-in-progress: BPNet: Branch-pruned conditional neural network for systematic time-accuracy tradeoff in DNN inference",

abstract = "Recently, there have been attempts to execute the neural network conditionally with auxiliary classifiers allowing early termination depending on the difficulty of the input, which can reduce the execution time or energy consumption without any or with negligible accuracy decrease. However, these studies do not consider how many or where the auxiliary classifiers, or branches, should be added in a systematic fashion. In this paper, we propose Branch-pruned Conditional Neural Network (BPNet) and its methodology in which the time-accuracy tradeoff for the conditional neural network can be found systematically. We applied BPNet to SqueezeNet, ResNet-20, and VGG-16 with CIFAR-10 and 100. BPNet achieves on average 2.0× of speedups without any accuracy drop on average compared to the base network.",

author = "Kyungchul Park and Youngmin Yi",

note = "Publisher Copyright: {\textcopyright} 2019 Association for Computing Machinery.; 2019 International Conference on Hardware/Software Codesign and System Synthesis, CODES/ISSS 2019 ; Conference date: 13-10-2019 Through 18-10-2019",

year = "2019",

month = oct,

day = "13",

doi = "10.1145/3349567.3351721",

language = "English",

series = "Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019",

publisher = "Association for Computing Machinery, Inc",

booktitle = "Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019",

}

Park, K & Yi, Y 2019, Work-in-progress: BPNet: Branch-pruned conditional neural network for systematic time-accuracy tradeoff in DNN inference. in Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019., a2, Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019, Association for Computing Machinery, Inc, 2019 International Conference on Hardware/Software Codesign and System Synthesis, CODES/ISSS 2019, New York, United States, 13/10/19. https://doi.org/10.1145/3349567.3351721

Work-in-progress: BPNet: Branch-pruned conditional neural network for systematic time-accuracy tradeoff in DNN inference. / Park, Kyungchul; Yi, Youngmin.
Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019. Association for Computing Machinery, Inc, 2019. a2 (Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - Work-in-progress

T2 - 2019 International Conference on Hardware/Software Codesign and System Synthesis, CODES/ISSS 2019

AU - Park, Kyungchul

AU - Yi, Youngmin

PY - 2019/10/13

Y1 - 2019/10/13

N2 - Recently, there have been attempts to execute the neural network conditionally with auxiliary classifiers allowing early termination depending on the difficulty of the input, which can reduce the execution time or energy consumption without any or with negligible accuracy decrease. However, these studies do not consider how many or where the auxiliary classifiers, or branches, should be added in a systematic fashion. In this paper, we propose Branch-pruned Conditional Neural Network (BPNet) and its methodology in which the time-accuracy tradeoff for the conditional neural network can be found systematically. We applied BPNet to SqueezeNet, ResNet-20, and VGG-16 with CIFAR-10 and 100. BPNet achieves on average 2.0× of speedups without any accuracy drop on average compared to the base network.

AB - Recently, there have been attempts to execute the neural network conditionally with auxiliary classifiers allowing early termination depending on the difficulty of the input, which can reduce the execution time or energy consumption without any or with negligible accuracy decrease. However, these studies do not consider how many or where the auxiliary classifiers, or branches, should be added in a systematic fashion. In this paper, we propose Branch-pruned Conditional Neural Network (BPNet) and its methodology in which the time-accuracy tradeoff for the conditional neural network can be found systematically. We applied BPNet to SqueezeNet, ResNet-20, and VGG-16 with CIFAR-10 and 100. BPNet achieves on average 2.0× of speedups without any accuracy drop on average compared to the base network.

UR - https://www.scopus.com/pages/publications/85077340365

U2 - 10.1145/3349567.3351721

DO - 10.1145/3349567.3351721

M3 - Conference contribution

AN - SCOPUS:85077340365

T3 - Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019

BT - Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019

PB - Association for Computing Machinery, Inc

Y2 - 13 October 2019 through 18 October 2019

ER -

Park K, Yi Y. Work-in-progress: BPNet: Branch-pruned conditional neural network for systematic time-accuracy tradeoff in DNN inference. In Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019. Association for Computing Machinery, Inc. 2019. a2. (Proceedings of the International Conference on Hardware/Software Codesign and System Synthesis Companion, CODES/ISSS 2019). doi: 10.1145/3349567.3351721

Work-in-progress: BPNet: Branch-pruned conditional neural network for systematic time-accuracy tradeoff in DNN inference

Abstract

Publication series

Conference

UN SDGs

Access to Document

Other files and links

Fingerprint

Cite this