An online generalized eigenvalue version of Laplacian Eigenmaps for visual big data

This paper presents a novel online version of laplacian eigenmap termed as generalized incremental laplacian eigenmap (GENILE), one of the most popular manifold-based dimensionality reduction technique performed by solving the generalized eigenvalue problem. We have used swiss roll and s-curve dataset, the most popular datasets used for manifold-based learning techniques, in this paper as artificial datasets. For a real data experiment, we have selected the MNIST digit dataset, the bank-note dataset, and the heart disease dataset for testing and evaluating our novel method and for comparing it with the standard batch isomap method and other manifold-based learning techniques. The experimental results have clearly shown the improvements in terms of classification accuracy by the proposed method in comparison with other techniques.