Group Differentiable Architecture Search

被引：0

作者：

Shen, Chaoyuan ^{[1
]}

Xu, Jinhua ^{[1
,2
]}

机构：

[1] East China Normal Univ, Schl Comp Sci & Technol, Shanghai 200062, Peoples R China

[2] East China Normal Univ, Shanghai Key Lab Multidimens Informat Proc, Shanghai 200062, Peoples R China

来源：

IEEE ACCESS | 2021年 / 9卷

基金：

中国国家自然科学基金;

关键词：

Computer architecture; Microprocessors; Training; Search problems; Optimization; Limiting; Licenses; Differentiable architecture search; skip-connect aggregation; shallow architecture;

D O I：

10.1109/ACCESS.2021.3081579

中图分类号：

TP [自动化技术、计算机技术];

学科分类号：

0812 ;

摘要：

Differentiable architecture search (DARTS) has received great attention due to its simplicity and efficiency. However, there are two annoying problems. One is that searched architecture of normal cell tends to be shallow. The other is skip-connect aggregation caused by the unfair competition between operations. We find that fewer operations per edge is helpful to search for deeper architectures, so we divide the operations into groups and train these groups in turn. To explore competitiveness among all the operations in the search space, candidate operations will be regrouped in each epoch. In addition, the random grouping prevents the overfitting of the super network, and consequently avoids the skip-connect aggregation. We named this method GroupDARTS and evaluated these searched architectures, achieving a state-of-the-art result of 97.68% on CIFAR10 and a top-1 accuray of 75.5% on ImageNet.

引用

页码：76585 / 76591

页数：7

共 27 条

[11]

Liu HaiJing Liu HaiJing, 2018, The Proceedings of the Fifteenth Congress of China Sheep Industry Development Sponsored by the China Animal Husbandry Association in 2018, Henan, China, 10-11 October, 2018, P13

[12]

Liu H, 2019, PROCEEDINGS OF THE THIRD INTERNATIONAL SYMPOSIUM - EDUCATIONAL RESEARCH AND EDUCATIONAL TECHNOLOGY, 2019, P3

[13]

Luo Renqian., 2018, NIPS

[14] ShuffleNet V2: Practical Guidelines for Efficient CNN Architecture Design [J].

Ma, Ningning ;

Zhang, Xiangyu ;

Zheng, Hai-Tao ;

Sun, Jian .

COMPUTER VISION - ECCV 2018, PT XIV, 2018, 11218 :122-138

[15]

Pham H, 2018, PR MACH LEARN RES, V80

[16]

Real E, 2019, AAAI CONF ARTIF INTE, P4780

[17] MobileNetV2: Inverted Residuals and Linear Bottlenecks [J].

Sandler, Mark ;

Howard, Andrew ;

Zhu, Menglong ;

Zhmoginov, Andrey ;

Chen, Liang-Chieh .

2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2018, :4510-4520

[18]

Szegedy C, 2015, PROC CVPR IEEE, P1, DOI 10.1109/CVPR.2015.7298594

[19]

Tan MX, 2019, PROC CVPR IEEE, P2815, DOI [arXiv:1807.11626, 10.1109/CVPR.2019.00293]

[20] Fair DARTS: Eliminating Unfair Advantages in Differentiable Architecture Search [J].

Chu, Xiangxiang ;

Zhou, Tianbao ;

Zhang, Bo ;

Li, Jixiang .

COMPUTER VISION - ECCV 2020, PT XV, 2020, 12360 :465-480

← 1 2 3 →