Data Selection Using Genetic Algorithm to Improve Transfer Learning Efficiency in Brain-Machine-Interface Systems

Data selection has a great effect on the performance of trained Convolutional Neural Networks (CNNs) for Brain Machine Interface systems. In this paper, we combine transfer learning and Genetic Algorithms to optimize the training data of CNNs. We implement transfer learning between different subjects. The data selected by GA to improve the performance of the CNN after transfer learning can be used to compare the similarity of the brain activity between subjects. The results show that brain data selected by GA improves the CNN recognition accuracy and reduces the training time. In addition, the performance of trained CNN using transfer learning on different days shows that although the performance deteriorates still it is acceptable.