Estimation de densité sur sous-variété

A broad guiding principle in applied statistics is that high-dimensionnal data live on smaller dimensionnal structures. In this context, we investigate the problem of density estimation when the data are supported on an unknown submanifold M of possibly unknown dimension d. We will try to adapt standard nonparametric tools such as kernel methods to this framework, and discuss the effect of the lack of knowledge on M on the accuracy of the estimate.