A review of neural architecture search

被引：64
作者：
Baymurzina, Dilyara [1 ]
Golikov, Eugene [1 ,3 ]
Burtsev, Mikhail [1 ,2 ]
机构：
[1] Moscow Inst Phys & Technol, Neural Networks & Deep Learning Lab, 9 Inst Skiy Lane, Dolgoprudnyi 141701, Moscow Region, Russia
[2] Artificial Intelligence Res Inst AIRI, Moscow, Russia
[3] Ecole Polytech Fed Lausanne, Route Cantonale, CH-1015 Lausanne, Vaud, Switzerland
来源：
NEUROCOMPUTING | 2022年 / 474卷
关键词：
Neural architecture search; NAS; AutoML; EVOLUTION; NETWORKS;
D O I：
10.1016/j.neucom.2021.12.014
中图分类号：
TP18 [人工智能理论];
学科分类号：
081104 ; 0812 ; 0835 ; 1405 ;
摘要：
Despite the impressive progress in neural network architecture design, improving the performance of the existing state-of-the-art models has become increasingly challenging. For this reason, the paradigm for neural architecture design is shifting from being expert-driven to almost fully automated. An emerging body of research related to such machine-aided design is called a Neural Architecture Search (NAS). This paper reviews the recent works on NAS and highlights several crucial concepts and problems of this field. (c) 2021 Elsevier B.V. All rights reserved.
引用
收藏
页码：82 / 93
页数：12
相关论文
共 86 条
  • [51] Rawal A, 2020, ARXIV200513092
  • [52] Real E., 2017, PR MACH LEARN RES, P2902
  • [53] Real E, 2019, AAAI CONF ARTIF INTE, P4780
  • [54] Rechenberg I, 1973, EVOLUTIONSSTRATEGIE
  • [55] Risi S, 2012, ARTIF LIFE, V18, P331, DOI [10.1162/artl_a_00071, 10.1162/ARTL_a_00071]
  • [56] Risi S, 2010, LECT NOTES ARTIF INT, V6226, P533, DOI 10.1007/978-3-642-15193-4_50
  • [57] Ru Binxin, 2020, ARXIV200604492
  • [58] Salimans T., 2017, ICLR, P1
  • [59] MobileNetV2: Inverted Residuals and Linear Bottlenecks
    Sandler, Mark
    Howard, Andrew
    Zhu, Menglong
    Zhmoginov, Andrey
    Chen, Liang-Chieh
    [J]. 2018 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR), 2018, : 4510 - 4520
  • [60] Simonyan K, 2015, Arxiv, DOI arXiv:1409.1556
123456789