Reliable Network Search Based on Evolutionary Algorithm

In this paper, we propose a neural architecture compression method based on network search to design a lightweight model for the network compression. The reasonable search method is designed based on evolutionary algorithm and search space for searching the efficient neural architecture, called EANet. The experimental results on several benchmarks datasets show that the performance of the EANet is better and the storage space is smaller. Besides, the light-weight SSD variant detection network based on EANet is applied to a railway intelligent surveillance system, which achieves the practical application based on the CNN model.