Orthogonal matching pursuit-based incremental locally linear embedding algorithm

Abstract

Locally linear embedding (LLE) is one of powerful manifold learning algorithms. However, when new data are available, it is necessary for LLE to run again with both the new data and the original data together. Thus, several incremental methods have been proposed for LLE to solve this problem. Linear incremental method is currently the most commonly used incremental approach. But the pseudo inverse solution does not possibly exist when handling incremental process. This paper deals with an incremental LLE based on orthogonal matching pursuit (ILLE-OMP), which remedies the drawback. ILLE-OMP is also a linear incremental method. Compared with other linear incremental methods, experimental results show that ILLE-OMP is promising.