Hierarchical memory-constrained operator scheduling of neural architecture search networks

Zihan Wang; Chengcheng Wan; Yuting Chen; Ziyi Lin; He Jiang; Lei Qiao

doi:10.1145/3489517.3530472

Hierarchical memory-constrained operator scheduling of neural architecture search networks

Zihan Wang
, Chengcheng Wan
, Yuting Chen^*
, Ziyi Lin
, He Jiang
, Lei Qiao

^*此作品的通讯作者

Shanghai Jiao Tong University
The University of Chicago
Alibaba Group Holding Ltd.
Dalian University of Technology
CAS - Beijing Institute of Control Engineering

科研成果: 书/报告/会议事项章节 › 会议稿件 › 同行评审

14 引用（Scopus）

摘要

Neural Architecture Search (NAS) is widely used in industry, searching for neural networks meeting task requirements. Meanwhile, it faces a challenge in scheduling networks satisfying memory constraints. This paper proposes HMCOS that performs hierarchical memory-constrained operator scheduling of NAS networks: given a network, HMCOS constructs a hierarchical computation graph and employs an iterative scheduling algorithm to progressively reduce peak memory footprints. We evaluate HMCOS against RPO and Serenity (two popular scheduling techniques). The results show that HMCOS outperforms existing techniques in supporting more NAS networks, reducing 8.7∼42.4% of peak memory footprints, and achieving 137 - 283x of speedups in scheduling.

源语言	英语
主期刊名	Proceedings of the 59th ACM/IEEE Design Automation Conference, DAC 2022
出版商	Institute of Electrical and Electronics Engineers Inc.
页	493-498
页数	6
ISBN（电子版）	9781450391429
DOI	https://doi.org/10.1145/3489517.3530472
出版状态	已出版 - 10 7月 2022
已对外发布	是
活动	59th ACM/IEEE Design Automation Conference, DAC 2022 - San Francisco, 美国期限: 10 7月 2022 → 14 7月 2022

出版系列

姓名	Proceedings - Design Automation Conference
ISSN（印刷版）	0738-100X

会议

会议	59th ACM/IEEE Design Automation Conference, DAC 2022
国家/地区	美国
市	San Francisco
时期	10/07/22 → 14/07/22

访问文件

10.1145/3489517.3530472

其它文件与链接

链接到 Scopus 的出版物

引用此

Wang, Z., Wan, C., Chen, Y., Lin, Z., Jiang, H., & Qiao, L. (2022). Hierarchical memory-constrained operator scheduling of neural architecture search networks. 在 Proceedings of the 59th ACM/IEEE Design Automation Conference, DAC 2022 (页码 493-498). (Proceedings - Design Automation Conference). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1145/3489517.3530472

@inproceedings{d574f39c7cf248efb52341c810b02596,

title = "Hierarchical memory-constrained operator scheduling of neural architecture search networks",

abstract = "Neural Architecture Search (NAS) is widely used in industry, searching for neural networks meeting task requirements. Meanwhile, it faces a challenge in scheduling networks satisfying memory constraints. This paper proposes HMCOS that performs hierarchical memory-constrained operator scheduling of NAS networks: given a network, HMCOS constructs a hierarchical computation graph and employs an iterative scheduling algorithm to progressively reduce peak memory footprints. We evaluate HMCOS against RPO and Serenity (two popular scheduling techniques). The results show that HMCOS outperforms existing techniques in supporting more NAS networks, reducing 8.7∼42.4\% of peak memory footprints, and achieving 137 - 283x of speedups in scheduling.",

author = "Zihan Wang and Chengcheng Wan and Yuting Chen and Ziyi Lin and He Jiang and Lei Qiao",

note = "Publisher Copyright: {\textcopyright} 2022 ACM.; 59th ACM/IEEE Design Automation Conference, DAC 2022 ; Conference date: 10-07-2022 Through 14-07-2022",

year = "2022",

month = jul,

day = "10",

doi = "10.1145/3489517.3530472",

language = "英语",

series = "Proceedings - Design Automation Conference",

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

pages = "493--498",

booktitle = "Proceedings of the 59th ACM/IEEE Design Automation Conference, DAC 2022",

address = "美国",

}

Wang, Z, Wan, C, Chen, Y, Lin, Z, Jiang, H & Qiao, L 2022, Hierarchical memory-constrained operator scheduling of neural architecture search networks. 在 Proceedings of the 59th ACM/IEEE Design Automation Conference, DAC 2022. Proceedings - Design Automation Conference, Institute of Electrical and Electronics Engineers Inc., 页码 493-498, 59th ACM/IEEE Design Automation Conference, DAC 2022, San Francisco, 美国, 10/07/22. https://doi.org/10.1145/3489517.3530472

Hierarchical memory-constrained operator scheduling of neural architecture search networks. / Wang, Zihan; Wan, Chengcheng; Chen, Yuting 等.
Proceedings of the 59th ACM/IEEE Design Automation Conference, DAC 2022. Institute of Electrical and Electronics Engineers Inc., 2022. 页码 493-498 (Proceedings - Design Automation Conference).

科研成果: 书/报告/会议事项章节 › 会议稿件 › 同行评审

TY - GEN

T1 - Hierarchical memory-constrained operator scheduling of neural architecture search networks

AU - Wang, Zihan

AU - Wan, Chengcheng

AU - Chen, Yuting

AU - Lin, Ziyi

AU - Jiang, He

AU - Qiao, Lei

PY - 2022/7/10

Y1 - 2022/7/10

N2 - Neural Architecture Search (NAS) is widely used in industry, searching for neural networks meeting task requirements. Meanwhile, it faces a challenge in scheduling networks satisfying memory constraints. This paper proposes HMCOS that performs hierarchical memory-constrained operator scheduling of NAS networks: given a network, HMCOS constructs a hierarchical computation graph and employs an iterative scheduling algorithm to progressively reduce peak memory footprints. We evaluate HMCOS against RPO and Serenity (two popular scheduling techniques). The results show that HMCOS outperforms existing techniques in supporting more NAS networks, reducing 8.7∼42.4% of peak memory footprints, and achieving 137 - 283x of speedups in scheduling.

AB - Neural Architecture Search (NAS) is widely used in industry, searching for neural networks meeting task requirements. Meanwhile, it faces a challenge in scheduling networks satisfying memory constraints. This paper proposes HMCOS that performs hierarchical memory-constrained operator scheduling of NAS networks: given a network, HMCOS constructs a hierarchical computation graph and employs an iterative scheduling algorithm to progressively reduce peak memory footprints. We evaluate HMCOS against RPO and Serenity (two popular scheduling techniques). The results show that HMCOS outperforms existing techniques in supporting more NAS networks, reducing 8.7∼42.4% of peak memory footprints, and achieving 137 - 283x of speedups in scheduling.

UR - https://www.scopus.com/pages/publications/85137537213

U2 - 10.1145/3489517.3530472

DO - 10.1145/3489517.3530472

M3 - 会议稿件

AN - SCOPUS:85137537213

T3 - Proceedings - Design Automation Conference

SP - 493

EP - 498

BT - Proceedings of the 59th ACM/IEEE Design Automation Conference, DAC 2022

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 59th ACM/IEEE Design Automation Conference, DAC 2022

Y2 - 10 July 2022 through 14 July 2022

ER -

Hierarchical memory-constrained operator scheduling of neural architecture search networks

摘要

出版系列

会议

访问文件

其它文件与链接

指纹

引用此