TY - GEN
T1 - DeepN-JPEG
T2 - 55th Annual Design Automation Conference, DAC 2018
AU - Liu, Zihao
AU - Liu, Tao
AU - Wen, Wujie
AU - Jiang, Lei
AU - Xu, Jie
AU - Wang, Yanzhi
AU - Quan, Gang
N1 - Publisher Copyright:
© 2018 Association for Computing Machinery.
PY - 2018/6/24
Y1 - 2018/6/24
N2 - As one of most fascinating machine learning techniques, deep neural network (DNN) has demonstrated excellent performance in various intelligent tasks such as image classiication. DNN achieves such performance, to a large extent, by performing expensive training over huge volumes of training data. To reduce the data storage and transfer overhead in smart resource-limited Internet-of-Thing (IoT) systems, efective data compression is a "must-have" feature before transferring real-time produced dataset for training or classiication. While there have been many well-known image compression approaches (such as JPEG), we for the irst time ind that a human-visual based image compression approach such as JPEG compression is not an optimized solution for DNN systems, especially with high compression ratios. To this end, we develop an image compression framework tailored for DNN applications, named "DeepN-JPEG", to embrace the nature of deep cascaded information process mechanism of DNN architecture. Extensive experiments, based on "ImageNet" dataset with various state-ofthe- A rt DNNs, show that "DeepN-JPEG" can achieve ∼ 3.5× higher compression rate over the popular JPEG solution while maintaining the same accuracy level for image recognition, demonstrating its great potential of storage and power eiciency in DNN-based smart IoT system design.
AB - As one of most fascinating machine learning techniques, deep neural network (DNN) has demonstrated excellent performance in various intelligent tasks such as image classiication. DNN achieves such performance, to a large extent, by performing expensive training over huge volumes of training data. To reduce the data storage and transfer overhead in smart resource-limited Internet-of-Thing (IoT) systems, efective data compression is a "must-have" feature before transferring real-time produced dataset for training or classiication. While there have been many well-known image compression approaches (such as JPEG), we for the irst time ind that a human-visual based image compression approach such as JPEG compression is not an optimized solution for DNN systems, especially with high compression ratios. To this end, we develop an image compression framework tailored for DNN applications, named "DeepN-JPEG", to embrace the nature of deep cascaded information process mechanism of DNN architecture. Extensive experiments, based on "ImageNet" dataset with various state-ofthe- A rt DNNs, show that "DeepN-JPEG" can achieve ∼ 3.5× higher compression rate over the popular JPEG solution while maintaining the same accuracy level for image recognition, demonstrating its great potential of storage and power eiciency in DNN-based smart IoT system design.
UR - http://www.scopus.com/inward/record.url?scp=85053680439&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85053680439&partnerID=8YFLogxK
U2 - 10.1145/3195970.3196022
DO - 10.1145/3195970.3196022
M3 - Conference contribution
AN - SCOPUS:85053680439
SN - 9781450357005
T3 - Proceedings - Design Automation Conference
BT - Proceedings of the 55th Annual Design Automation Conference, DAC 2018
PB - Institute of Electrical and Electronics Engineers Inc.
Y2 - 24 June 2018 through 29 June 2018
ER -