Personalized adaptive instruction design (PAID) for brain-computer interface using reinforcement learning and deep learning: simulated data study

Brain-computer interface (BCI) systems may require the user to perform a set of mental tasks, such as imagining different types of motion. The performance demonstrated on these tasks varies with time and between users. This study presents a new method for the automatically adaptive, user-specific generation of a sequence of tasks to increase the effectiveness of user training. For this purpose, we developed the Personalized Adaptive Instruction Design (PAID) algorithm, which uses reinforcement learning and deep learning. Using simulated data, we compared the training strategy developed here with uniform random and sequential selection strategies. The results demonstrate that the PAID strategy outperforms the others and is close to the theoretically optimal solution. Moreover, our algorithm offers the possibility of efficiently integrating psychological aspects of the training process into the generated strategy.