Unsupervised Learning of Manifolds via Linear Approximations

In this paper, we examine the application of manifold learning to the clustering problem. The method used is Locality Preserving Projections (LPP), which is chosen because of its computational efficiency. A detailed derivation of the method is presented, as well as the theoretical justification behind it. Experiments performed on CMU's PIE database show that the projections created by LPP yield better clustering results than those obtained by k-means alone.