Structural Laplacian Eigenmaps for Modeling Sets of Multivariate Sequences

What is it about?

Dimensionality reduction is the transformation or/and combination of the original multidimensional features in order to generate more informative, descriptive and practical data representation in a space of fewer dimensions. This process is achieved by eliminating redundancies and irrelevant relationships present in datasets while ensuring maximum preservation of the original information. These techniques have proved an essential step in many machine learning applications in domains such as computer vision. This paper offers a novel non-linear dimensionality reduction method dedicated to multivariate sequences.

Featured Image

Why is it important?

This presents a novel dimensionality reduction method (Structural Laplacian Eigenmaps) that explicitly deals with the problem of dimensionality reduction of time series and clearly outperforms any previous approaches. Results are presented on many datasets of high-dimensional time series, such as 3D motion capture sequences and view-variant sets of object images. The preceding conference paper’s (ICPR2010) high citations suggest there is a significant interest in this research.