Enhance cross-subject MEG decoding by combining Riemannian Geometry feature extraction and multi-task learning

Many brain decoding algorithms are based on the assumption that the training and test data are in the same feature space and have the same distribution. However, in many practical applications, the statistical distribution of brain signal varies across subjects as well as sessions, restricting the transferability of training data or training models between different subjects or sessions. In this study, we focus on this problem in magnetoencephalography (MEG) signal. By reviewing the transfer learning literatures that have achieved satisfactory results in brain decoding, we proposed a new method for MEG decoding across subjects. First, a new kernel is derived by establishing a connection between the Riemannian geometry and the sample covariance matrix (SCM), which is then applied to MEG transfer decoding by using an improved multi-task learning framework. The experiments on an MEG dataset of face vs. scramble decoding task demonstrated that the proposed approach is superior to other comparable methods.