Small hypersphere fitting by Spherical Least Squares Jun Fujiki Abstract: To measure the similarity between two high dimensional vector data, the correlation coefficient is often used instead of Euclidean distance, that is, high dimensional vectors are normalized as hyperspherical points. In this paper, we propose the methods fitting a low dimensional small hypersphere to high dimensional data on the unit hypersphere by the spherical least squares. We also evaluate the methods by synthesized data.