Some aspects of deep representation learning on transformed EEG data

Abstract

Visualizing high-dimensional datasets can be challenging. While it is possible to plot data in two or three dimensions to reveal the data's innate structure, analogous high-dimensional representations are significantly less understandable. A dataset's structure must be shown to some extent, hence the dimension must be decreased. Principal component analysis (PCA) and linear discriminant analysis (LDA) were the two historically the first methods. Several nonlinear techniques were afterwards developed, including locally linear embedding (LLE), multi-dimensional scaling (MDS), isometric feature mapping (Isomap), stochastic neighborhood embedding (t-SNE), etc. In the current study, several nonlinear representation learning techniques are used for electroencephalography (EEG) data with the ultimate objective of categorizing the EEG signal.