A critical aspect of non-linear dimensionality reduction techniques is represented by the construction of the adjacency graph. The difficulty resides in finding the optimal parameters, a process which, in general, is heuristically driven. Recently, sparse representation has been proposed as a non-parametric solution to overcome this problem. In this paper, we demonstrate that this approach not only serves for the graph construction, but also represents an efficient and accurate alternative for out-of-sample embedding. Considering for a case study the Laplacian Eigenmaps, we applied our method to the face recognition problem. Experimental results conducted on some challenging datasets confirmed the robustness of our approach and its superiority when compared to existing techniques. © 2012 Springer-Verlag Berlin Heidelberg.
CITATION STYLE
Raducanu, B., & Dornaika, F. (2012). Out-of-sample embedding by sparse representation. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 7626 LNCS, pp. 336–344). https://doi.org/10.1007/978-3-642-34166-3_37
Mendeley helps you to discover research relevant for your work.