Maintaining Orientation of a Local Neighborhood in High-dimensional Space for Real-time Visualization of Massive Data

Abstract

Navigation in high-dimensional space in an efficient and intuitive manner has been a challenging task. This paper introduces a visualization technique that extracts a local neighborhood and aligns it so that the directions are maintained while zooming in to the local neighborhood from a global view, performing dimensionality reduction such as PCA (principal components analysis) for the local neighborhood at the same time. Extraction of a local neighborhood allows us to examine the data in real-time, without data points far away overlapping each other as is observed in conventional dimensionality reduction methods.