Transfer Learning for the Riemannian Tangent Space: Applications to Brain-Computer Interfaces
Alexandre Bleuzé, Jérémie Mattout, Marco Congedo
Abstract
Transfer learning for training brain-computer interface (BCI) decoding algorithms is useful to reduce the calibration time, increase the accuracy, reduce the risk of overfitting and allow the application of machine learning methods that require a large amount of data, such as deep neuronal networks. In this article we propose a transfer learning method inspired by recent advances in Riemannian geometry. The method aligns vectors in the tangent space of a source and a target data set by means of Procrustes Analysis. We apply the method on a publicly available P300-BCI database. We show that using our method it is possible to transfer information reusing data from other subjects. The classification accuracy we obtain, as compared to the state of art, shows a clear transmission of information using the transfer learning method.