Abstract：The structure of the convolutional neural network will also affect its performance. The design of the convolutional neural network relies more on experience and powerful computing power. How to design a neural network with better performance lacks effective theoretical support. In order to solve this problem, based on the analysis of the complexity of the typical convolutional neural network topology, in order to quickly realize the convolutional neural network that meets the given complexity characteristics, the generation from complex network topology to convolutional neural network is given. The algorithm, through the establishment of a series of convolutional neural networks with different topological features, uses the Cifar10 and Cifar100 data sets to analyze the relationship between the average clustering coefficient, average path length, graph density, modularity and other topological properties on the recognition effectiveness of the convolutional neural network. Experiments show that when the number of parameters of the neural network is basically equal, the average clustering coefficient will affect the performance of the convolutional neural network. The final conclusion is that in a statistical sense, a network structure with a small average clustering coefficient will have better performance, which provides a theoretical basis for further designing a better convolutional neural network.
王光波, 孙仁诚, 隋毅, 邵峰晶. 卷积神经网络复杂性质与准确率的关系研究[J]. 复杂系统与复杂性科学, 2021, 18(2): 60-65.
WANG Guangbo, SUN Rencheng, SUI Yi, SHAO Fengjing. On the Relationship Between the Complexity and Accuracy of Convolutional Neural Networks. Complex Systems and Complexity Science, 2021, 18(2): 60-65.
[1]Simard D, Nadeau L, Kroger H.Faster learning in small-world neural networks[J].Physics Letters A, 2005, 336(1): 8-15. [2]Xie S, Kirillov A, Girshick R, et al. Exploring randomly wired neural networks for image recognition[DB/OL].[2020-03-20]. https://arxiv.org/pdf/1904.01569.pdf. [3]Lécun Y, Bottou L. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11):2278-2324. [4]Alom M Z, Taha T M, Yakopcic C, et al. The history began from alexnet: a comprehensive survey on deep learning approaches[DB/OL].[2020-03-20].https://arxiv.org/ftp/arxiv/papers/1803/1803.01164.pdf. [5]Simonyan K, Zisserman A. Very deep convolutional networks for large-scale image recognition[DB/OL].[2020-03-20]. https://arxiv.org/pdf/1409.1556v6.pdf. [6]Szegedy C, Ioffe S, Vanhoucke V, et al. Inception-v4, inception-resnet and the impact of residual connections on learning[DB/OL].[2020-03-20].https://arxiv.org/pdf/1602.07261v1.pdf. [7]He K, Zhang X, Ren S, et al. Deep residual learning for image recognition[DB/OL].[2020-03-20]. https://arxiv.org/pdf/1512.03385.pdf. [8]Huang G, Liu Z, Van Der Maaten L, et al. Densely connected convolutional networks[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Honolulu, HI, USA,2017: 4700-4708. [9]Zoph B, Vasudevan V, Shlens J, et al. Learning transferable architectures for scalable image recognition[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, UT, USA, 2018: 8697-8710. [10] Eggemann N, Noble S D. The clustering coefficient of a scale-free random graph[J]. Discrete Applied Mathematics, 2011, 159(10): 953-965. [11] Newman M E. The structure and function of complex networks[J].Siam Review,2003,45(2): 167-256. [12] Newman M E J, Girvan M. Finding and evaluating community structure in networks[J]. Physical Review E, 2004, 69(2): 026113. [13] Ioffe S, Szegedy C. Batch normalization: accelerating deep network training by reducing internal covariate shift[C]//International Conference on Machine Learning. Miami, Florida, USA, 2015: 448-456. [14] Xu L, Choy C, Li Y W. Deep sparse rectifier neural networks for speech denoising[C]//2016 IEEE International Workshop on Acoustic Signal Enhancement. Xi'an,China, 2016: 1-5. [15] Kingma D P, Ba L J. Adam: a method for stochastic optimization[J]. [DB/OL].[2020-03-20]. https://arxiv.org/pdf/1412.6980v9.pdf.
