Genetic Algorithm Based Deep Learning Neural Network Structure and Hyperparameter Optimization

Alzheimer’s disease is one of the major challenges population ageing, and diagnosis prediction through various biomarkers key. While application deep learning as imaging technologies has recently expanded across medical industry, empirical design these very difficult. The main reason for this problem that performance Convolutional Neural Networks (CNN) differ greatly depending on statistical distribution input dataset. Different hyperparameters also affect convergence CNN models. With amount information, selecting appropriate parameters network structure became a large research area. Genetic Algorithm (GA), popular technique to automatically select high-performance architecture. In paper, we show possibility optimising architecture using GA, where its search space includes both configuration hyperparameters. To verify our Algorithm, used an amyloid brain image dataset diagnosis. As result, algorithm outperforms by 11.73% given classification task.