Surrogate-Assisted Multiobjective Neural Architecture Search for Real-Time Semantic Segmentation

作     者：Lu, Zhichao Cheng, Ran Huang, Shihua Zhang, Haoming Qiu, Changxiao Yang, Fan 

作者机构：Southern University of Science and Technology Guangdong Key Laboratory of Brain-Inspired Intelligent Computation Department of Computer Science and Engineering Shenzhen518055 China Huawei Technologies Co. Ltd. Hisilicon Research Department Shenzhen518055 China 

出 版 物：《IEEE Transactions on Artificial Intelligence》 (IEEE. Trans. Artif. Intell.)

年 卷 期：2023年第4卷第6期

页      面：1602-1615页

核心收录：

基　　金：National Natural Science Foundation of China China Postdoctoral Science Foundation Guangdong Provincial Key Laboratory 

主　　题：Computer architecture 

摘      要：The architectural advancements in deep neural networks have led to remarkable leap-forwards across a broad array of computer vision tasks. Instead of relying on human expertise, neural architecture search (NAS) has emerged as a promising avenue toward automating the design of architectures. While recent achievements on image classification have suggested opportunities, the promises of NAS have yet to be thoroughly assessed on more challenging tasks of semantic segmentation. The main challenges of applying NAS to semantic segmentation arise from two aspects: 1) high-resolution images to be processed;2) additional requirement of real-time inference speed (i.e., real-time semantic segmentation) for applications such as autonomous driving. To meet such challenges, we propose a surrogate-assisted multiobjective method in this article. Through a series of customized prediction models, our method effectively transforms the original NAS task to an ordinary multiobjective optimization problem. Followed by a hierarchical prescreening criterion for in-fill selection, our method progressively achieves a set of efficient architectures trading-off between segmentation accuracy and inference speed. Empirical evaluations on three benchmark datasets together with an application using Huawei Atlas 200 DK suggest that our method can identify architectures significantly outperforming existing state-of-the-art architectures designed both manually by human experts and automatically by other NAS methods. Code is available from here. © 2020 IEEE.