Asymmetric Isomap for Dimensionality Reduction and Data Visualization

We propose an asymmetric version of the Isomap dimensionality reduction and data visualization approach. Our improvement uses the information on asymmetric input data relationships, and in this way, it determines the input dissimilarities more accurately than original Isomap. We introduce as well the asymmetric coefficients discovering and expressing the asymmetric properties of the input data. These coefficients asymmetrize geodesic distances in Isomap making this method asymmetric. The experiments on two real datasets confirm the effectiveness of our proposal.