基于神经进化的深度学习模型研究综述

doi:10.12263/DZXB.20200139

摘要/Abstract

摘要： 深度学习研究发展至今已可以胜任各类识别、分类、生成任务，但是对于不同的任务，神经网络的结构或参数不可能只是微小的变化，依然需要专家进行调整.在这样的情况下，自动化地调整神经网络的结构或参数成为研究热点.其中，以达尔文自然进化论为灵感的神经进化成为主要优化方法.利用神经进化优化的深度学习模型以种群为基础，通过突变、重组等操作进化，可实现自动地、逐步地构建神经网络并最终选择出性能最优的深度学习模型.本文简述了神经进化与进化计算；详细概述了各类基于神经进化的深度学习模型；分析了各类模型的性能；总结了神经进化与深度学习融合的前景并探讨下一步的研究方向.

关键词: 神经进化, 深度学习, 进化计算, 卷积神经网络, 生成式对抗网络, 自动编码器, 长短期记忆网络, 深度强化学习

Abstract: With the development of deep learning,it has been competent to various types of tasks,such as identification,classification,and generation.However,the structures or parameters of artificial neural networks cannot be only a little changed for different task,experts need to adjust the structures or parameters of the neural network.In such situations,the method of automatically adjusting the structures or parameters of the artificial neural network has become a research hotspot,among these methods,neuroevolution inspired by Darwin's natural evolution theory has become the main optimization method for that.Deep learning models optimized by neuroevolution based on population,evolving through mutation,crossover and other operations,can automatically and gradually construct the neural network and then choose the most optimal deep learning model.This paper summarizes the neuroevolution and the evolutionary computation.It elaborates various deep learning models based on neuroevolution,and analyzes the performance of these models.It concludes prospects of the deep learning model based on neuroevolution and discusses the next research directions.

Key words: neuroevolution, deep learning, evolutionary computation, convolutional neural networks, generative adversarial networks, autoencoder, long short-term memory networks, deep reinforcement learning

中图分类号:

TP391.4

韩冲, 王俊丽, 吴雨茜, 张超波. 基于神经进化的深度学习模型研究综述[J]. 电子学报, 2021, 49(2): 372-379.

HAN Chong, WANG Jun-li, WU Yu-xi, ZHANG Chao-bo. A Review of Deep Learning Models Based on Neuroevolution[J]. Acta Electronica Sinica, 2021, 49(2): 372-379.

参考文献

[1] Ketkar N.Deep Learning with Python[M].Berkeley:Apress,2017.113-132.
[2] Floreano D,et al.Neuroevolution:from architectures to learning[J].Evolutionary Intelligence,2008,1(1):47-62.
[3] Fogel D B,Fogel L J.An introduction to evolutionary programming[A].European Conference on Artificial Evolution[C]. Berlin:Springer,1995.21-33.
[4] Andrew A M.Systems:an introductory analysis with applications to biology,control,and artificial intelligence[J].Robotica,1993,11(5):489-489.
[5] Beyer H G,Schwefel H P.Evolution strategies-a comprehensive introduction[J].Natural Computing,2002,1(1):3-52.
[6] Koza J R,Koza J R.Genetic Programming:On the Programming of Computers by Means of Natural Selection[M].Cambridge:MIT Press,1992.
[7] 贾亚军,丛爽.进化算法的分析及对比研究[A].2009系统仿真技术及其应用学术会议论文集[C].合肥:中国科学技术大学出版社,2009.855-861.
[8] Zhang C,Lim P,Qin A K,et al.Multiobjective deep belief networks ensemble for remaining useful life estimation in prognostics[J].IEEE Transactions on Neural Networks and Learning Systems,2016,28(10):2306-2318.
[9] Smith C,Jin Y.Evolutionary multi-objective generation of recurrent neural network ensembles for time series prediction[J].Neurocomputing,2014,143(1):302-311.
[10] Lukoševi Ač ius M,Jaeger H.Reservoir computing approaches to recurrent neural network training[J].Computer Science Review,2009,3(3):127-149.
[11] Evans B,Al-Sahaf H,Xue B,et al.Evolutionary deep learning:a genetic programming approach to image classification[A].IEEE Congress on Evolutionary Computation[C].New York,USA:IEEE,2018.1-6.
[12] Atkins D,Neshatian K,Zhang M.A domain independent genetic programming approach to automatic feature extraction for image classification[A].IEEE Congress of Evolutionary Computation[C].New York,USA:IEEE,2011.238-245.
[13] Suganuma M,Shirakawa S,Nagao T.A genetic programming approach to designing convolutional neural network architectures[A].Proceedings of the Genetic and Evolutionary Computation Conference[C].New York,USA:ACM,2017.497-504.
[14] Xie L,Yuille A.Genetic CNN[A].Proceedings of the IEEE International Conference on Computer Vision[C].New York,USA:IEEE,2017.1379-1388.
[15] Simonyan K,Zisserman A.Very Deep Convolutional Networks for Large-scale Image Recognition[EB/OL].heeps:arXiv.org/abs/1409.1556,2014.
[16] He K,Zhang X,Ren S,et al.Deep residual learning for image recognition[A].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C].New York,USA:IEEE,2016.770-778.
[17] Huang G,Liu Z,Van Der Maaten L,et al.Densely connected convolutional networks[A].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C].New York,USA:IEEE,2017.4700-4708.
[18] Wang C,Xu C,Yao X,et al.Evolutionary generative adversarial networks[J].IEEE Transactions on Evolutionary Computation,2019,23(6):921-934.
[19] Goodfellow I,Pouget-Abadie J,Mirza M,et al.Generative adversarial nets[A].Advances in Neural Information Processing Systems[C].Cambridge,UK:MIT Press,2014.2672-2680.
[20] Lin J.Divergence measures based on the shannon entropy[J].IEEE Transactions on Information Theory,1991,37(1):145-151.
[21] Assuncao F,Sereno D,Lourenco N,et al.Automatic evolution of autoencoders for compressed representations[A].IEEE Congress on Evolutionary Computation[C].New York,USA:IEEE,2018.1-8.
[22] Suganuma M,Ozay M,Okatani T.Exploiting the Potential of Standard Convolutional Autoencoders for Image Restoration by Evolutionary Search[EB/OL].https:arXiv.org/abs/1803.00370,2018.
[23] Miikkulainen R,Liang J,Meyerson E,et al.Artificial Intelligence in the Age of Neural Networks and Brain Computing[M].Amsterdam:Elsevier Inc,2018.293-312.
[24] Stanley K O,Miikkulainen R.Evolving neural networks through augmenting topologies[J].Evolutionary Computation,2002,10(2):99-127.
[25] Agrawal R K,Muchahary F,Tripathi M M.Long term load forecasting with hourly predictions based on long-short-term-memory networks[A].2018 IEEE Texas Power and Energy Conference[C].New York,USA:IEEE,2018.1-6.
[26] Mnih V,Kavukcuoglu K,Silver D,et al.Playing Atari with Deep Reinforcement Learning[EB/OL].https:arXiv.org/abs/1312.5602,2013.
[27] Salimans T,Ho J,Chen X,et al.Evolution Strategies as a Scalable Alternative to Reinforcement Learning[EB/OL].https:arXiv.org/abs/1703.03864,2017.
[28] Such F P,Madhavan V,Conti E,et al.Deep Neuroevolution:Genetic Algorithms are a Competitive Alternative for Training Deep Neural Networks for Reinforcement Learning[EB/OL].https:arXiv.org/abs/1712.06567,2017.
[29] Lehman J,Chen J,Clune J,et al.ES is more than just a traditional finite-difference approximator[A].Proceedings of the Genetic and Evolutionary Computation Conference[C].New York,USA:ACM,2018.450-457.
[30] Conti E,Madhavan V,Such F P,et al.Improving exploration in evolution strategies for deep reinforcement learning via a population of novelty-seeking agents[A].Advances in Neural Information Processing Systems[C].Combridge:MIT Press,2018.5027-5038.
[31] Gaier A,Ha D.Weight agnostic neural networks[A].Advances in Neural Information Processing Systems[C]. Combridge,UK:MIT Press,2019.5365-5379.
[32] Miller B L,Goldberg D E.Genetic algorithms,tournament selection,and the effects of noise[J].Complex Systems,1995,9(3):193-212.
[33] Morse G,Stanley K O.Simple evolutionary optimization can rival stochastic gradient descent in neural networks[A].Proceedings of the Genetic and Evolutionary Computation Conference[C].New York,USA:Association Computing Machinery,2016.477-484.
[34] Krizhevsky A,Sutskever I,Hinton G E.Imagenet classification with deep convolutional neural networks[A].Advances in Neural Information Processing Systems[C].New York,USA:Association Computing Machinery,2012.1097-1105.
[35] Szegedy C,Liu W,Jia Y,et al.Going deeper with convolutions[A].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C].New York,USA:IEEE,2015.1-9.
[36] Sara U,Akter M,Uddin M S.Image quality assessment through FSIM,SSIM,MSE and PSNR-A comparative study[J].Journal of Computer and Communications,2019,7(3):8-18.
[37] Krause J,Stark M,Deng J,et al.3D object representations for fine-grained categorization[A].Proceedings of the IEEE International Conference on Computer Vision Workshops[C].New York,USA:IEEE, 2013.554-561.
[38] Pathak D,Krahenbuhl P,Donahue J,et al.Context encoders:feature learning by inpainting[A].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C].New York,USA:IEEE,2016.2536-2544.
[39] Yeh R A,Chen C,Yian Lim T,et al.Semantic image inpainting with deep generative models[A].Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition[C].New York,USA:IEEE,2017.5485-5493.