References & Citations
Mathematics > Optimization and Control
Title: Riemannian preconditioned coordinate descent for low multi-linear rank approximation
(Submitted on 3 Sep 2021 (this version), latest version 21 Mar 2024 (v2))
Abstract: This paper presents a fast, memory efficient, optimization-based, first-order method for low multi-linear rank approximation of high-order, high-dimensional tensors.
In our method, we exploit the second-order information of the cost function and the constraints to suggest a new Riemannian metric on the Grassmann manifold.
We use a Riemmanian coordinate descent method for solving the problem, and also provide a local convergence analysis matching that of the coordinate descent method in the Euclidean setting. We also show that each step of our method with unit step-size is actually a step of the orthogonal iteration algorithm. Experimental results show the computational advantage of our method for high-dimensional tensors.
Submission history
From: Reshad Hosseini [view email][v1] Fri, 3 Sep 2021 17:18:36 GMT (415kb,D)
[v2] Thu, 21 Mar 2024 13:44:02 GMT (391kb,D)
Link back to: arXiv, form interface, contact.