Micro Darknet for Inference: ESL reference for inference accelerator design

Min Zhi Ji; Wei Chung Tseng; Ting Jia Wu; Bo Rong Lin; Chung Ho Chen

doi:10.1109/ISOCC47750.2019.9027644

Micro Darknet for Inference: ESL reference for inference accelerator design

Min Zhi Ji, Wei Chung Tseng, Ting Jia Wu, Bo Rong Lin, Chung Ho Chen

Institute of Computer and Communication Engineering

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

1 Citation (Scopus)

Abstract

For neural network (NN) models applying to low-end edge devices, the memory management is a very important issue because of the limitation of hardware resources. However, current NN frameworks typically allocate a huge memory space for NN models in the initial stage. To reduce memory requirements, we propose a lite NN inference-only framework, MDFI (Micro Darknet for Inference) based on Darknet. We optimize the MDFI C code by a layer-wise memory management and layer-dependency resolving mechanism. According to the experimental results, the average memory consumption of MDFI has 76% reduction compared to Darknet, and the average execution time of MDFI has 8% reduction also.

Original language	English
Title of host publication	Proceedings - 2019 International SoC Design Conference, ISOCC 2019
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	69-70
Number of pages	2
ISBN (Electronic)	9781728124780
DOIs	https://doi.org/10.1109/ISOCC47750.2019.9027644
Publication status	Published - 2019 Oct
Event	16th International System-on-Chip Design Conference, ISOCC 2019 - Jeju, Korea, Republic of Duration: 2019 Oct 6 → 2019 Oct 9

Publication series

Name	Proceedings - 2019 International SoC Design Conference, ISOCC 2019

Conference

Conference	16th International System-on-Chip Design Conference, ISOCC 2019
Country/Territory	Korea, Republic of
City	Jeju
Period	19-10-06 → 19-10-09

All Science Journal Classification (ASJC) codes

Signal Processing
Electrical and Electronic Engineering
Instrumentation
Artificial Intelligence
Hardware and Architecture

Access to Document

10.1109/ISOCC47750.2019.9027644

Cite this

Ji, M. Z., Tseng, W. C., Wu, T. J., Lin, B. R., & Chen, C. H. (2019). Micro Darknet for Inference: ESL reference for inference accelerator design. In Proceedings - 2019 International SoC Design Conference, ISOCC 2019 (pp. 69-70). Article 9027644 (Proceedings - 2019 International SoC Design Conference, ISOCC 2019). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/ISOCC47750.2019.9027644

@inproceedings{388510746f6c4731bdf7532577004cb8,

title = "Micro Darknet for Inference: ESL reference for inference accelerator design",

abstract = "For neural network (NN) models applying to low-end edge devices, the memory management is a very important issue because of the limitation of hardware resources. However, current NN frameworks typically allocate a huge memory space for NN models in the initial stage. To reduce memory requirements, we propose a lite NN inference-only framework, MDFI (Micro Darknet for Inference) based on Darknet. We optimize the MDFI C code by a layer-wise memory management and layer-dependency resolving mechanism. According to the experimental results, the average memory consumption of MDFI has 76% reduction compared to Darknet, and the average execution time of MDFI has 8% reduction also.",

author = "Ji, {Min Zhi} and Tseng, {Wei Chung} and Wu, {Ting Jia} and Lin, {Bo Rong} and Chen, {Chung Ho}",

note = "Publisher Copyright: {\textcopyright} 2019 IEEE. Copyright: Copyright 2020 Elsevier B.V., All rights reserved.; 16th International System-on-Chip Design Conference, ISOCC 2019 ; Conference date: 06-10-2019 Through 09-10-2019",

year = "2019",

month = oct,

doi = "10.1109/ISOCC47750.2019.9027644",

language = "English",

series = "Proceedings - 2019 International SoC Design Conference, ISOCC 2019",

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

pages = "69--70",

booktitle = "Proceedings - 2019 International SoC Design Conference, ISOCC 2019",

address = "United States",

}

Ji, MZ, Tseng, WC, Wu, TJ, Lin, BR & Chen, CH 2019, Micro Darknet for Inference: ESL reference for inference accelerator design. in Proceedings - 2019 International SoC Design Conference, ISOCC 2019., 9027644, Proceedings - 2019 International SoC Design Conference, ISOCC 2019, Institute of Electrical and Electronics Engineers Inc., pp. 69-70, 16th International System-on-Chip Design Conference, ISOCC 2019, Jeju, Korea, Republic of, 19-10-06. https://doi.org/10.1109/ISOCC47750.2019.9027644

Micro Darknet for Inference: ESL reference for inference accelerator design. / Ji, Min Zhi; Tseng, Wei Chung; Wu, Ting Jia et al.
Proceedings - 2019 International SoC Design Conference, ISOCC 2019. Institute of Electrical and Electronics Engineers Inc., 2019. p. 69-70 9027644 (Proceedings - 2019 International SoC Design Conference, ISOCC 2019).

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

TY - GEN

T1 - Micro Darknet for Inference

T2 - 16th International System-on-Chip Design Conference, ISOCC 2019

AU - Ji, Min Zhi

AU - Tseng, Wei Chung

AU - Wu, Ting Jia

AU - Lin, Bo Rong

AU - Chen, Chung Ho

PY - 2019/10

Y1 - 2019/10

N2 - For neural network (NN) models applying to low-end edge devices, the memory management is a very important issue because of the limitation of hardware resources. However, current NN frameworks typically allocate a huge memory space for NN models in the initial stage. To reduce memory requirements, we propose a lite NN inference-only framework, MDFI (Micro Darknet for Inference) based on Darknet. We optimize the MDFI C code by a layer-wise memory management and layer-dependency resolving mechanism. According to the experimental results, the average memory consumption of MDFI has 76% reduction compared to Darknet, and the average execution time of MDFI has 8% reduction also.

AB - For neural network (NN) models applying to low-end edge devices, the memory management is a very important issue because of the limitation of hardware resources. However, current NN frameworks typically allocate a huge memory space for NN models in the initial stage. To reduce memory requirements, we propose a lite NN inference-only framework, MDFI (Micro Darknet for Inference) based on Darknet. We optimize the MDFI C code by a layer-wise memory management and layer-dependency resolving mechanism. According to the experimental results, the average memory consumption of MDFI has 76% reduction compared to Darknet, and the average execution time of MDFI has 8% reduction also.

UR - http://www.scopus.com/inward/record.url?scp=85082995802&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85082995802&partnerID=8YFLogxK

U2 - 10.1109/ISOCC47750.2019.9027644

DO - 10.1109/ISOCC47750.2019.9027644

M3 - Conference contribution

AN - SCOPUS:85082995802

T3 - Proceedings - 2019 International SoC Design Conference, ISOCC 2019

SP - 69

EP - 70

BT - Proceedings - 2019 International SoC Design Conference, ISOCC 2019

PB - Institute of Electrical and Electronics Engineers Inc.

Y2 - 6 October 2019 through 9 October 2019

ER -

Ji MZ, Tseng WC, Wu TJ, Lin BR, Chen CH. Micro Darknet for Inference: ESL reference for inference accelerator design. In Proceedings - 2019 International SoC Design Conference, ISOCC 2019. Institute of Electrical and Electronics Engineers Inc. 2019. p. 69-70. 9027644. (Proceedings - 2019 International SoC Design Conference, ISOCC 2019). doi: 10.1109/ISOCC47750.2019.9027644

Micro Darknet for Inference: ESL reference for inference accelerator design

Abstract

Publication series

Conference

All Science Journal Classification (ASJC) codes

Access to Document

Other files and links

Fingerprint

Cite this