Codes and Expansions (CodEx) Seminar

Hau-Tieng Wu (Duke University)
Some recent progress in diffusion based manifold learning and its applications

Diffusion based machine learning algorithms have been actively developed and applied to analyze complicated datasets in past decades. However, there are still many interesting challenging problems left. I will share some recent progress, particularly the \(L^\infty\) spectral convergence and robustness to heterogeneous and colored noise under the manifold setup. Its application to spatiotemporal analysis for biomedical nonstationary signals will be demonstrated as a motivating example. If time permits, some analyses under the kernel random matrix setup with the bandwidth selection problem will be discussed.