A0749
Title: Intrinsic and extrinsic deep learning on manifolds
Authors: Lizhen Lin - The University of Notre Dame (United States) [presenting]
Abstract: Both intrinsic and extrinsic deep neural network (DNN) models on the manifolds are discussed. An intrinsic DNN employs a Riemannian structure of the manifold while an extrinsic DNN relies on embedding a manifold onto a high-dimensional Euclidean space. The excessive risk of the DNN estimators will be derived and extensive numerical studies have been carried out to demonstrate the utilities of the models and illustrate the role of the geometry in developing the DNN models.
