EMONAS-Net: Efficient multiobjective neural architecture search using surrogate-assisted evolutionary algorithm for 3D medical image segmentation

作者：

Highlights：

• An efficient multiobjective neural architecture search framework is presented.

• The search space considers the micro and macro structure of the architecture.

• The proposed surrogate-assisted evolutionary algorithm improves convergence.

• The framework was tested on three 3D medical image segmentation challenges.

• The proposed framework efficiently identifies accurate and smaller architectures.

摘要

•An efficient multiobjective neural architecture search framework is presented.•The search space considers the micro and macro structure of the architecture.•The proposed surrogate-assisted evolutionary algorithm improves convergence.•The framework was tested on three 3D medical image segmentation challenges.•The proposed framework efficiently identifies accurate and smaller architectures.

论文关键词：Medical image segmentation,Convolutional neural networks,Neural architecture search,Hyperparameter optimization,AutoML,Multiobjective optimization

论文评审过程：Received 27 August 2020, Revised 17 June 2021, Accepted 16 August 2021, Available online 24 August 2021, Version of Record 4 September 2021.

论文官网地址：https://doi.org/10.1016/j.artmed.2021.102154