Abstract

This workshop aims to bring together researchers with different expertise to exchange ideas on designing computational methods for various tensor problems. One of the main features will be a focus on rigorous algorithms for which one may guarantee convergence to a global solution under reasonable conditions. Topics will include provable algorithmic guarantees in terms of computational efficiency and robustness, algorithmic techniques like spectral methods, iterative methods for nonconvex optimization, and powerful convex relaxation hierarchies like sum-of-squares, etc, where correctness guarantees may be established under suitable assumptions motivated by applications. A secondary goal is to better delineate and understand the boundary separating the possible from the impossible in tensor computations — computational tractability vs intractability, existence/uniqueness vs nonexistence/nonuniqueness of solutions, tameness vs wildness, etc.