Genetic Algorithm Based Optimization of Deep Neural Network Ensemble for Personal Identification in Pedestrians Behaviors
Xuanang Feng, Jianing Zhao, Eisuke Kita · 2019
Personal identification is a task of authenticating a person using individual biological features. Deep neural networks (DNNs) have demonstrated impressive performance in this field. It is well known, however, that no general algorithm is available for every application problem. For a new application task, it is very time-consuming for non-experts to design network structure, hyperparameters and an ensemble of base models adequately and effectively. In this paper, we present the genetic algorithm (GA) based approach in order to construct network structures automatically, tune their hyperparameters adequately and generate base models for the ensemble algorithm. Then the ensemble is constructed from the base models with different network structures. Our original personal identification dataset is employed as the numerical example in order to illustrate the performance of the proposed method. Convergence property, experiment results, as well as search performance are discussed in the numerical results. The results indicate that several single base models provided by the proposed algorithm have a strong classification ability by exploring the features of a pedestrian's behaviors and that an ensemble model which is constituted from these base models achieves better classification performance.