DANCE: Differentiable Accelerator/Network Co-Exploration

Kanghyun Choi; Deokki Hong; Hojae Yoon; Joonsang Yu; Youngsok Kim; Jinho Lee

doi:10.1109/DAC18074.2021.9586121

DANCE: Differentiable Accelerator/Network Co-Exploration

Kanghyun Choi, Deokki Hong, Hojae Yoon, Joonsang Yu, Youngsok Kim, Jinho Lee

College of Computing

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

7 Citations (Scopus)

Abstract

This work presents DANCE, a differentiable approach towards the co-exploration of hardware accelerator and network architecture design. At the heart of DANCE is a differentiable evaluator network. By modeling the hardware evaluation software with a neural network, the relation between the accelerator design and the hardware metrics becomes differentiable, allowing the search to be performed with backpropagation. Compared to the naive existing approaches, our method performs co-exploration in a significantly shorter time, while achieving superior accuracy and hardware cost metrics.

Original language	English
Title of host publication	2021 58th ACM/IEEE Design Automation Conference, DAC 2021
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	337-342
Number of pages	6
ISBN (Electronic)	9781665432740
DOIs	https://doi.org/10.1109/DAC18074.2021.9586121
Publication status	Published - 2021 Dec 5
Event	58th ACM/IEEE Design Automation Conference, DAC 2021 - San Francisco, United States Duration: 2021 Dec 5 → 2021 Dec 9

Publication series

Name	Proceedings - Design Automation Conference
Volume	2021-December
ISSN (Print)	0738-100X

Conference

Conference	58th ACM/IEEE Design Automation Conference, DAC 2021
Country/Territory	United States
City	San Francisco
Period	21/12/5 → 21/12/9

Bibliographical note

Funding Information:
ACKNOWLEDGMENT This work was supported by National Research Foundation of Korea (NRF) grant funded by the Korea government(MSIT) (No.2020R1F1A1074472) and Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No.2020-0-01361, Artificial Intelligence Graduate School Program (Yonsei University)) and the Yonsei University Research Fund (2020-22-0512)

Publisher Copyright:
© 2021 IEEE.

All Science Journal Classification (ASJC) codes

Computer Science Applications
Control and Systems Engineering
Electrical and Electronic Engineering
Modelling and Simulation

Access to Document

10.1109/DAC18074.2021.9586121

Cite this

@inproceedings{0041dc8e1eba4150b08362300d2f9f5c,

title = "DANCE: Differentiable Accelerator/Network Co-Exploration",

abstract = "This work presents DANCE, a differentiable approach towards the co-exploration of hardware accelerator and network architecture design. At the heart of DANCE is a differentiable evaluator network. By modeling the hardware evaluation software with a neural network, the relation between the accelerator design and the hardware metrics becomes differentiable, allowing the search to be performed with backpropagation. Compared to the naive existing approaches, our method performs co-exploration in a significantly shorter time, while achieving superior accuracy and hardware cost metrics.",

author = "Kanghyun Choi and Deokki Hong and Hojae Yoon and Joonsang Yu and Youngsok Kim and Jinho Lee",

note = "Funding Information: ACKNOWLEDGMENT This work was supported by National Research Foundation of Korea (NRF) grant funded by the Korea government(MSIT) (No.2020R1F1A1074472) and Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No.2020-0-01361, Artificial Intelligence Graduate School Program (Yonsei University)) and the Yonsei University Research Fund (2020-22-0512) Publisher Copyright: {\textcopyright} 2021 IEEE.; 58th ACM/IEEE Design Automation Conference, DAC 2021 ; Conference date: 05-12-2021 Through 09-12-2021",

year = "2021",

month = dec,

day = "5",

doi = "10.1109/DAC18074.2021.9586121",

language = "English",

series = "Proceedings - Design Automation Conference",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "337--342",

booktitle = "2021 58th ACM/IEEE Design Automation Conference, DAC 2021",

address = "United States",

}

Choi, K, Hong, D, Yoon, H, Yu, J, Kim, Y & Lee, J 2021, DANCE: Differentiable Accelerator/Network Co-Exploration. in 2021 58th ACM/IEEE Design Automation Conference, DAC 2021. Proceedings - Design Automation Conference, vol. 2021-December, Institute of Electrical and Electronics Engineers Inc., pp. 337-342, 58th ACM/IEEE Design Automation Conference, DAC 2021, San Francisco, United States, 21/12/5. https://doi.org/10.1109/DAC18074.2021.9586121

DANCE : Differentiable Accelerator/Network Co-Exploration. / Choi, Kanghyun; Hong, Deokki; Yoon, Hojae et al.

2021 58th ACM/IEEE Design Automation Conference, DAC 2021. Institute of Electrical and Electronics Engineers Inc., 2021. p. 337-342 (Proceedings - Design Automation Conference; Vol. 2021-December).

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

TY - GEN

T1 - DANCE

T2 - 58th ACM/IEEE Design Automation Conference, DAC 2021

AU - Choi, Kanghyun

AU - Hong, Deokki

AU - Yoon, Hojae

AU - Yu, Joonsang

AU - Kim, Youngsok

AU - Lee, Jinho

N1 - Funding Information: ACKNOWLEDGMENT This work was supported by National Research Foundation of Korea (NRF) grant funded by the Korea government(MSIT) (No.2020R1F1A1074472) and Institute of Information & communications Technology Planning & Evaluation (IITP) grant funded by the Korea government (MSIT) (No.2020-0-01361, Artificial Intelligence Graduate School Program (Yonsei University)) and the Yonsei University Research Fund (2020-22-0512) Publisher Copyright: © 2021 IEEE.

PY - 2021/12/5

Y1 - 2021/12/5

N2 - This work presents DANCE, a differentiable approach towards the co-exploration of hardware accelerator and network architecture design. At the heart of DANCE is a differentiable evaluator network. By modeling the hardware evaluation software with a neural network, the relation between the accelerator design and the hardware metrics becomes differentiable, allowing the search to be performed with backpropagation. Compared to the naive existing approaches, our method performs co-exploration in a significantly shorter time, while achieving superior accuracy and hardware cost metrics.

AB - This work presents DANCE, a differentiable approach towards the co-exploration of hardware accelerator and network architecture design. At the heart of DANCE is a differentiable evaluator network. By modeling the hardware evaluation software with a neural network, the relation between the accelerator design and the hardware metrics becomes differentiable, allowing the search to be performed with backpropagation. Compared to the naive existing approaches, our method performs co-exploration in a significantly shorter time, while achieving superior accuracy and hardware cost metrics.

UR - http://www.scopus.com/inward/record.url?scp=85119401786&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85119401786&partnerID=8YFLogxK

U2 - 10.1109/DAC18074.2021.9586121

DO - 10.1109/DAC18074.2021.9586121

M3 - Conference contribution

AN - SCOPUS:85119401786

T3 - Proceedings - Design Automation Conference

SP - 337

EP - 342

BT - 2021 58th ACM/IEEE Design Automation Conference, DAC 2021

PB - Institute of Electrical and Electronics Engineers Inc.

Y2 - 5 December 2021 through 9 December 2021

ER -

DANCE: Differentiable Accelerator/Network Co-Exploration

Abstract

Publication series

Conference

Bibliographical note

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this