Joint Approximate Diagonalization under Orthogonality Constraints (2110.03235v1)

Abstract: Joint diagonalization of a set of positive (semi)-definite matrices has a wide range of analytical applications, such as estimation of common principal components, estimation of multiple variance components, and blind signal separation. However, when the eigenvectors of involved matrices are not the same, joint diagonalization is a computationally challenging problem. To the best of our knowledge, currently existing methods require at least $O(KN^3)$ time per iteration, when $K$ different $N \times N$ matrices are considered. We reformulate this optimization problem by applying orthogonality constraints and dimensionality reduction techniques. In doing so, we reduce the computational complexity for joint diagonalization to $O(N^3)$ per quasi-Newton iteration. This approach we refer to as JADOC: Joint Approximate Diagonalization under Orthogonality Constraints. We compare our algorithm to two important existing methods and show JADOC has superior runtime while yielding a highly similar degree of diagonalization. The JADOC algorithm is implemented as open-source Python code, available at https://github.com/devlaming/jadoc.