论文部分内容阅读
We present a continuum perspective on several nonlinear data analysis methods,with a special focus on Eigenmaps and Diffusion Maps.We derive bounds on the complexity of the continuous versions of these algorithms,in terms of the geometry of the manifold they are trying to learn.In particular,we bound the number of eigenfunctions needed to accurately represent the manifold in a(low-dimensional)Euclidean space.