Abstract An algorithm for neural network structure design is proposed. The algorithm introduces indirect encoding schema to represent the structure of neural network and the cell in the 2-dimension cellular automate system to represent the existence of connection in neural network. By separately evolving the coordinate and value of the cell, the growing and pruning of the network structure are achieved. The coordinate of the cell is created and evolved by binary quantum particle swarm optimization (BQPSO). The value of the cell is evolved by using properly-designed neighboring evolving rule of cellular system, and the current network is trained by float-point QPSO. Thus, the final stable structure is found. The experimental results show that the proposed algorithm has stable complexity and convergent capability with different scales of neural network structure design.
[1] Miller G F, Todd P, Hegde S. Designing Neural Networks Using Genetic Algorithm // Proc of the 3rd International Conference on Genetic Algorithm. Fairfax County, USA, 1989: 379-384 [2] Merril J W L, Port R F. Fractally Configured Neural Network. Neural Networks, 1991, 4(1): 53-60 [3] Sun Jun, Xu Wenbo, Fang Wei, et al. Quantum-Behaved Particle Swarm Optimization with Binary Encoding // Proc of the 8th International Conference on Adaptive and Natural Computing Algorithms. Warsaw, Poland, 2007: 376-385 [4] Zhang Chuanwu. Progress of Cellular Automata and Its Theory Study. Journal of Guizhou University: Natural Sciences, 2004, 21(3): 289-292,306 (in Chinese) (张传武.细胞自动机及其理论研究进展.贵州大学学报:自然科学版, 2004, 21(3): 289-292,306) [5] Cao Xingqin, Wang Nengchao. A Novel Parameterization for the Space of Cellular Automata Rules. Computer Science, 2007, 34(3): 145-147,164 (in Chinese) (曹兴芹,王能超.新型细胞自动机规则空间的参数化. 计算机科学, 2007,34(3):145-147,164) [6] Sun Jun, Feng Bin, Xu Wenbo. Particle Swarm Optimization with Particles Having Quantum Behavior // Proc of the Congress on Evolutionary Computation. San Diego, USA, 2004: 325-331 [7] Kennedy J, Eberhart R. Particle Swarm Optimization // Proc of the IEEE International Conference on Neural Network. Perth, Australia, 1995: 1942-1948 [8] Chen Wei, Feng Bin, Sun Jun. Simulation Study on the Parameters Optimization of Radial Basis Function Neural Network Based on QPSO Algorithm. Journal of Computer Applications, 2006, 26(8): 1928-1931 (in Chinese) (陈 伟,冯 斌,孙 俊.基于QPSO算法的RBF神经网络参数优化仿真研究.计算机应用, 2006, 26(8): 1928-1931) [9] Bao Fang, Pan Yonghui, Xu Wenbo. A Novel Training Algorithm for BP Neural Network // Proc of the International Symposium on Distributed Computing and Application to Business, Engineering and Science. Hangzhou, China, 2006: 767-770 [10] Soniag E D. Feedforward Nets for Interpolation and Classification. Journal of Computer and System Science, 1992, 45(1): 20-48
2009, Vol. 22 
