Automatic Dimensionality Estimation for Manifold Learning through Optimal Feature Selection
A very important aspect in manifold learning is represented by automatic estimation of the intrinsic dimensionality. Unfortunately, this problem has received few attention in the literature of manifold learning. In this paper, we argue that feature selection paradigm can be used to the problem of automatic dimensionality estimation. Besides this, it also leads to improved recognition rates. Our approach for optimal feature selection is based on a Genetic Algorithm. As a case study for manifold learning, we have considered Laplacian Eigenmaps (LE) and Locally Linear Embedding (LLE). The effectiveness of the proposed framework was tested on the face recognition problem. Extensive experiments carried out on ORL, UMIST, Yale, and Extended Yale face data sets confirmed our hypothesis.
KeywordsFeature Selection Recognition Rate Feature Selection Algorithm Locally Linear Embedding Manifold Learn
- 2.Saul, L.K., Roweis, S.T., Singer, Y.: Think globally, fit locally: Unsupervised learning of low dimensional manifolds. Journal of Machine Learning Research 4, 119–155 (2003)Google Scholar
- 9.Zhao, Z., Liu, H.: Spectral feature selection for supervised and unsupervised learning. In: Int. Conference on Machine Learning (2007)Google Scholar
- 11.He, X., Cai, D., Niyogi, P.: Laplacian score for feature selection. In: Advances in Neural Information Processing Systems 18 (2005)Google Scholar
- 12.Cai, D., Zhang, C., He, X.: Unsupervised feature selection for multi-cluster data. In: 16th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, KDD 2010 (2010)Google Scholar
- 13.Liu, H., Yu, L.: Toward integrating feature selection algorithms for classification and clustering. IEEE Trans. Knowledge Data Engineering 17, 494–502 (2005)Google Scholar