Statistical Learning for Brain-Computer Interfaces

This chapter introduces statistical learning and its applications to brain-computer interfaces (BCIs). It presents the general principles of supervised learning and discusses the difficulties raised by its implementation, with a particular focus on aspects related to selecting sensors and multisubject learning. The chapter also describes how a learning approach may be validated, including various metrics of performance and optimization of the hyperparameters of the considered algorithms. The goal of supervised learning is to construct a predictor function that assigns a label to any given example; this predictor function is constructed from labeled examples that provide a basis for this training process. One of the possible approaches for building BCIs that require less calibration with new users is to use training techniques based on information transfer, or multitask training techniques. Validating the results obtained in a given application serves two purposes in statistical learning: evaluating the chosen performance metric and optimizing the hyperparameters of the algorithm.