Michael Margaliot
Tel Aviv University, Israel
1556-603X/21©2021IEEE AUGUST 2021 | IEEE COMPUTATIONAL INTELLIGENCE MAGAZINE 67
Abstract—Recently, convolutional neural networks (CNNs)
have achieved great success in the field of artificial intelligence,
including speech recognition, image recognition, and natural
language processing. CNN architecture plays a key role in
CNNs’ performance. Most previous CNN architectures are
hand-crafted, which requires designers to have rich expert
domain knowledge. The trial-and-error process consumes a lot
of time and computing resources. To solve this problem,
researchers proposed the neural architecture search, which
searches CNN architecture automatically, to satisfy dif-
ferent requirements. However, the blindness of the search
strategy causes a ‘loss of experience’ in the early stage of
the search process, and ultimately affects the results of the
later stage. In this paper, we propose a self-adaptive
mutation neural architecture search algorithm based on
ResNet blocks and DenseNet blocks. The self-adaptive
mutation strategy makes the algorithm adaptively adjust
the mutation strategies during the evolution process to
achieve better exploration. In addition, the whole search pro-
cess is fully automatic, and users do not need expert knowl-
edge about CNNs architecture design. In this paper, the
proposed algorithm is compared with 17 state-of-the-art
algorithms, including manually designed CNN and automatic
search algorithms on CIFAR10 and CIFAR100. The results
indicate that the proposed algorithm outperforms the com-
petitors in terms of classification performance and consumes
fewer computing resources.
Digital Object Identifier 10.1109/MCI.2021.3084435
Date of current version: 15 July 2021 Corresponding Author: Yu Xue (e-mail: xueyu@nuist.edu.cn).
Yu Xue and Yankang Wang
Nanjing University of Information Science and Technology, CHINA
Jiayu Liang
Tiangong University, CHINA
Adam Slowik
Koszalin University of Technology, POLAND
A Self-Adaptive Mutation
Neural Architecture Search
Algorithm Based on Blocks
©SHUTTERSTOCK.COM/BETTERVECTOR
Authorized licensed use limited to: Nanjing University of Information Science and Technology. Downloaded on July 22,2021 at 05:28:02 UTC from IEEE Xplore. Restrictions apply.