|
Abstract:
|
In this talk we introduce a new method for performing joint dimension reduction, or manifold learning, along the modes of a data tensor, in order to re-order the fibers along each mode so that the resulting permuted tensor is smooth. Our approach generalizes recent work on a convex formulation of the co-clustering problem. Like convex co-clustering, our co-manifold learning procedure possesses stability guarantees with respect to perturbations in the input data tensor. We illustrate how our method can identify the coupled intrinsic geometries in simulated and real data examples.
|
