Global linear convergent algorithm to compute the minimum volume enclosing ellipsoid

TL;DR

This paper introduces a globally linear convergent first-order algorithm for the MVEE problem, demonstrating advantages in high-dimensional data and establishing connections between different optimization approaches.

Contribution

It proposes the ACD algorithm with proven global linear convergence and offers new insights into coordinate selection and algorithm relationships.

Findings

ACD algorithm is globally linear convergent.

ACD outperforms other algorithms in high-dimensional cases.

The paper proves the linear convergence of Wolfe-Atwood's algorithm.

Abstract

The minimum volume enclosing ellipsoid (MVEE) problem is an optimization problem in the basis of many practical problems. This paper describes some new properties of this model and proposes a first-order oracle algorithm, the Adjusted Coordinate Descent (ACD) algorithm, to address the MVEE problem. The ACD algorithm is globally linear convergent and has an overwhelming advantage over the other algorithms in cases where the dimension of the data is large. Moreover, as a byproduct of the convergence property of the ACD algorithm, we prove the global linear convergence of the Frank-Wolfe type algorithm (illustrated by the case of Wolfe-Atwood's algorithm), which supports the conjecture of Todd. Furthermore, we provide a new interpretation for the means of choosing the coordinate axis of the Frank-Wolfe type algorithm from the perspective of the smoothness of the coordinate axis, i.e., the algorithm chooses the coordinate axis with the worst smoothness at each iteration. This finding connects the first-order oracle algorithm and the linear optimization oracle algorithm on the MVEE problem. The numerical tests support our theoretical results.