TECO : A Unified Feature Map Compression Framework Based on Transform and Entropy
The massive memory accesses of feature maps (FMs) in deep neural network (DNN) processors lead to huge power consumption, which becomes a major energy bottleneck of DNN accelerators. In this article, we propose a unified framework named Transform and Entropy-based COmpression (TECO) scheme to efficiently compress FMs with various attributes in DNN inference. We explore, for the first time, the intrinsic unimodal distribution characteristic that widely exists in the frequency domain of various FMs. In addition, a well-optimized hardware-friendly coding scheme is designed, which fully utilizes this remarkable data distribution characteristic to encode and compress the frequency spectrum of different FMs. Furthermore, the information entropy theory is leveraged to develop a novel loss function for improving the compression ratio and to make a fast comparison among different compressors. Extensive experiments are performed on multiple tasks and demonstrate that the proposed TECO achieves compression ratios of 2.31 × in ResNet-50 on image classification, 3.47 × in UNet on dark image enhancement, and 3.18 × in Yolo-v4 on object detection while keeping the accuracy of these models. Compared with the upper limit of the compression ratio for original FMs, the proposed framework achieves the compression ratio improvement of 21%, 157%, and 152% on the above models.
Medienart: |
E-Artikel |
---|
Erscheinungsjahr: |
2023 |
---|---|
Erschienen: |
2023 |
Enthalten in: |
Zur Gesamtaufnahme - volume:PP |
---|---|
Enthalten in: |
IEEE transactions on neural networks and learning systems - PP(2023) vom: 13. Sept. |
Sprache: |
Englisch |
---|
Beteiligte Personen: |
Shi, Yubo [VerfasserIn] |
---|
Links: |
---|
Themen: |
---|
Anmerkungen: |
Date Revised 13.02.2024 published: Print-Electronic Citation Status Publisher |
---|
doi: |
10.1109/TNNLS.2023.3309667 |
---|
funding: |
|
---|---|
Förderinstitution / Projekttitel: |
|
PPN (Katalog-ID): |
NLM362004765 |
---|
LEADER | 01000caa a22002652 4500 | ||
---|---|---|---|
001 | NLM362004765 | ||
003 | DE-627 | ||
005 | 20240213232501.0 | ||
007 | cr uuu---uuuuu | ||
008 | 231226s2023 xx |||||o 00| ||eng c | ||
024 | 7 | |a 10.1109/TNNLS.2023.3309667 |2 doi | |
028 | 5 | 2 | |a pubmed24n1290.xml |
035 | |a (DE-627)NLM362004765 | ||
035 | |a (NLM)37703155 | ||
040 | |a DE-627 |b ger |c DE-627 |e rakwb | ||
041 | |a eng | ||
100 | 1 | |a Shi, Yubo |e verfasserin |4 aut | |
245 | 1 | 0 | |a TECO |b A Unified Feature Map Compression Framework Based on Transform and Entropy |
264 | 1 | |c 2023 | |
336 | |a Text |b txt |2 rdacontent | ||
337 | |a ƒaComputermedien |b c |2 rdamedia | ||
338 | |a ƒa Online-Ressource |b cr |2 rdacarrier | ||
500 | |a Date Revised 13.02.2024 | ||
500 | |a published: Print-Electronic | ||
500 | |a Citation Status Publisher | ||
520 | |a The massive memory accesses of feature maps (FMs) in deep neural network (DNN) processors lead to huge power consumption, which becomes a major energy bottleneck of DNN accelerators. In this article, we propose a unified framework named Transform and Entropy-based COmpression (TECO) scheme to efficiently compress FMs with various attributes in DNN inference. We explore, for the first time, the intrinsic unimodal distribution characteristic that widely exists in the frequency domain of various FMs. In addition, a well-optimized hardware-friendly coding scheme is designed, which fully utilizes this remarkable data distribution characteristic to encode and compress the frequency spectrum of different FMs. Furthermore, the information entropy theory is leveraged to develop a novel loss function for improving the compression ratio and to make a fast comparison among different compressors. Extensive experiments are performed on multiple tasks and demonstrate that the proposed TECO achieves compression ratios of 2.31 × in ResNet-50 on image classification, 3.47 × in UNet on dark image enhancement, and 3.18 × in Yolo-v4 on object detection while keeping the accuracy of these models. Compared with the upper limit of the compression ratio for original FMs, the proposed framework achieves the compression ratio improvement of 21%, 157%, and 152% on the above models | ||
650 | 4 | |a Journal Article | |
700 | 1 | |a Wang, Meiqi |e verfasserin |4 aut | |
700 | 1 | |a Cao, Tianyu |e verfasserin |4 aut | |
700 | 1 | |a Lin, Jun |e verfasserin |4 aut | |
700 | 1 | |a Wang, Zhongfeng |e verfasserin |4 aut | |
773 | 0 | 8 | |i Enthalten in |t IEEE transactions on neural networks and learning systems |d 2012 |g PP(2023) vom: 13. Sept. |w (DE-627)NLM23236897X |x 2162-2388 |7 nnns |
773 | 1 | 8 | |g volume:PP |g year:2023 |g day:13 |g month:09 |
856 | 4 | 0 | |u http://dx.doi.org/10.1109/TNNLS.2023.3309667 |3 Volltext |
912 | |a GBV_USEFLAG_A | ||
912 | |a GBV_NLM | ||
951 | |a AR | ||
952 | |d PP |j 2023 |b 13 |c 09 |