WorksComputer Science
-number and Its Applications to Data Science
Abstract
In this paper, we introduce a novel metric, -number, designed to measure the "-th order -dimensional oscillation degree" of a set. For a given set in the Euclidean space , -number is computed by identifying an optimal -dimensional plane that minimizes the average -th order distance of all points in to . This metric serves as a versatile tool in data science, functioning both as a dimensionality reduction technique to find the optimal -th order -dimensional subspace and as a measure to characterize data oscillation for classification. We conduct a thorough exploration of -number through theoretical analysis and numerical experiments. Initially, we establish the upper and lower bounds of -number for any set , along with its scaling law. We then elucidate the relationships between -number and its related parameters. A key finding of this study is that in discrete settings, when , the -dimensional plane determined by -number coincides with that obtained via Principal Component Analysis (PCA), positioning -number as a generalization of PCA. Furthermore, we propose two algorithms for calculating -number in both continuous and discrete settings. We validate our theoretical results through a series of numerical experiments in low-dimensional cases. Finally, we demonstrate the potential of -number in practical data analysis through a dimensionality reduction task and an image classification task.