SEFACT: Selective feature activation and early classification for CNNs

Farhana Sharmin Snigdha; Ibrahim Ahmed; Susmita Dey Manasi; Meghna G. Mankalale; Jiang Hu; Sachin S Sapatnekar

doi:10.1145/3287624.3287663

SEFACT: Selective feature activation and early classification for CNNs

Farhana Sharmin Snigdha, Ibrahim Ahmed, Susmita Dey Manasi, Meghna G. Mankalale, Jiang Hu, Sachin S Sapatnekar

Electrical and Computer Engineering

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

Abstract

This work presents a dynamic energy reduction approach for hardware accelerators for convolutional neural networks (CNN). Two methods are used: (1) an adaptive data-dependent scheme to selectively activate a subset of all neurons, by narrowing down the possible activated classes (2) static bitwidth reduction. The former is applied in late layers of the CNN, while the latter is more effective in early layers. Even accounting for the implementation overheads, the results show 20%-25% energy savings with 5-10% accuracy loss.

Original language	English (US)
Title of host publication	ASP-DAC 2019 - 24th Asia and South Pacific Design Automation Conference
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	526-531
Number of pages	6
ISBN (Electronic)	9781450360074
DOIs	https://doi.org/10.1145/3287624.3287663
State	Published - Jan 21 2019
Event	24th Asia and South Pacific Design Automation Conference, ASPDAC 2019 - Tokyo, Japan Duration: Jan 21 2019 → Jan 24 2019

Publication series

Name	Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC

Other

Other	24th Asia and South Pacific Design Automation Conference, ASPDAC 2019
Country	Japan
City	Tokyo
Period	1/21/19 → 1/24/19

Access

10.1145/3287624.3287663

Fingerprint Dive into the research topics of 'SEFACT: Selective feature activation and early classification for CNNs'. Together they form a unique fingerprint.

Cite this

Snigdha, F. S., Ahmed, I., Manasi, S. D., Mankalale, M. G., Hu, J., & Sapatnekar, S. S. (2019). SEFACT: Selective feature activation and early classification for CNNs. In ASP-DAC 2019 - 24th Asia and South Pacific Design Automation Conference (pp. 526-531). (Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1145/3287624.3287663

SEFACT : Selective feature activation and early classification for CNNs. / Snigdha, Farhana Sharmin; Ahmed, Ibrahim; Manasi, Susmita Dey; Mankalale, Meghna G.; Hu, Jiang; Sapatnekar, Sachin S.

ASP-DAC 2019 - 24th Asia and South Pacific Design Automation Conference. Institute of Electrical and Electronics Engineers Inc., 2019. p. 526-531 (Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC).

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

Snigdha, FS, Ahmed, I, Manasi, SD, Mankalale, MG, Hu, J & Sapatnekar, SS 2019, SEFACT: Selective feature activation and early classification for CNNs. in ASP-DAC 2019 - 24th Asia and South Pacific Design Automation Conference. Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC, Institute of Electrical and Electronics Engineers Inc., pp. 526-531, 24th Asia and South Pacific Design Automation Conference, ASPDAC 2019, Tokyo, Japan, 1/21/19. https://doi.org/10.1145/3287624.3287663

Snigdha FS, Ahmed I, Manasi SD, Mankalale MG, Hu J, Sapatnekar SS. SEFACT: Selective feature activation and early classification for CNNs. In ASP-DAC 2019 - 24th Asia and South Pacific Design Automation Conference. Institute of Electrical and Electronics Engineers Inc. 2019. p. 526-531. (Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC). https://doi.org/10.1145/3287624.3287663

Snigdha, Farhana Sharmin ; Ahmed, Ibrahim ; Manasi, Susmita Dey ; Mankalale, Meghna G. ; Hu, Jiang ; Sapatnekar, Sachin S. / SEFACT : Selective feature activation and early classification for CNNs. ASP-DAC 2019 - 24th Asia and South Pacific Design Automation Conference. Institute of Electrical and Electronics Engineers Inc., 2019. pp. 526-531 (Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC).

@inproceedings{3a86eeb6210c4b4a9fe66d180b451645,

title = "SEFACT: Selective feature activation and early classification for CNNs",

abstract = "This work presents a dynamic energy reduction approach for hardware accelerators for convolutional neural networks (CNN). Two methods are used: (1) an adaptive data-dependent scheme to selectively activate a subset of all neurons, by narrowing down the possible activated classes (2) static bitwidth reduction. The former is applied in late layers of the CNN, while the latter is more effective in early layers. Even accounting for the implementation overheads, the results show 20%-25% energy savings with 5-10% accuracy loss.",

author = "Snigdha, {Farhana Sharmin} and Ibrahim Ahmed and Manasi, {Susmita Dey} and Mankalale, {Meghna G.} and Jiang Hu and Sapatnekar, {Sachin S}",

note = "Publisher Copyright: {\textcopyright} 2019 Copyright is held by the owner/author(s). Publication rights licensed to ACM. Copyright: Copyright 2019 Elsevier B.V., All rights reserved.; 24th Asia and South Pacific Design Automation Conference, ASPDAC 2019 ; Conference date: 21-01-2019 Through 24-01-2019",

year = "2019",

month = jan,

day = "21",

doi = "10.1145/3287624.3287663",

language = "English (US)",

series = "Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC",

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

pages = "526--531",

booktitle = "ASP-DAC 2019 - 24th Asia and South Pacific Design Automation Conference",

}

TY - GEN

T1 - SEFACT

T2 - 24th Asia and South Pacific Design Automation Conference, ASPDAC 2019

AU - Snigdha, Farhana Sharmin

AU - Ahmed, Ibrahim

AU - Manasi, Susmita Dey

AU - Mankalale, Meghna G.

AU - Hu, Jiang

AU - Sapatnekar, Sachin S

PY - 2019/1/21

Y1 - 2019/1/21

N2 - This work presents a dynamic energy reduction approach for hardware accelerators for convolutional neural networks (CNN). Two methods are used: (1) an adaptive data-dependent scheme to selectively activate a subset of all neurons, by narrowing down the possible activated classes (2) static bitwidth reduction. The former is applied in late layers of the CNN, while the latter is more effective in early layers. Even accounting for the implementation overheads, the results show 20%-25% energy savings with 5-10% accuracy loss.

AB - This work presents a dynamic energy reduction approach for hardware accelerators for convolutional neural networks (CNN). Two methods are used: (1) an adaptive data-dependent scheme to selectively activate a subset of all neurons, by narrowing down the possible activated classes (2) static bitwidth reduction. The former is applied in late layers of the CNN, while the latter is more effective in early layers. Even accounting for the implementation overheads, the results show 20%-25% energy savings with 5-10% accuracy loss.

UR - http://www.scopus.com/inward/record.url?scp=85061132645&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85061132645&partnerID=8YFLogxK

U2 - 10.1145/3287624.3287663

DO - 10.1145/3287624.3287663

M3 - Conference contribution

AN - SCOPUS:85061132645

T3 - Proceedings of the Asia and South Pacific Design Automation Conference, ASP-DAC

SP - 526

EP - 531

BT - ASP-DAC 2019 - 24th Asia and South Pacific Design Automation Conference

PB - Institute of Electrical and Electronics Engineers Inc.

Y2 - 21 January 2019 through 24 January 2019

ER -