Abstract:Because designed by hand plus limited times of trial and error,traditional deep convolutional neural network design methods usually result in excessively redundant parameters and huge multiplication times.To automatically construct deep convolutional neural networks with features like flexible structures,small-scale,and low multiplication times,this paper proposes a multi-objective neural evolution algorithm for deep convolutional networks.The algorithm expresses a deep neural network as a directed graph and uses neural evolution and multi-objective optimization algorithm to achieve simultaneous multi-objective optimization of depth,computational costs,and recognition accuracy.The method uses linear programming to translate genetic codes into convolutional layers to allow the evolutionary algorithm to automatically adjust the specific configuration of each network layer.The evolved model has 36 convolution layers and Top5 accuracy 86.1%,Top1 accuracy 60.2%on CIFAR-100.Compared with networks with similar recognition accuracy,the evolved network has more novel structures and fewer multiplication times.In summary,the proposed approach can automatically create a series of deep neural networks with different features.It is a fast,economical,and automated design method for industrial applications,especially resource-limited applications.