Splitting numerical integration for matrix completion
This work addresses matrix completion for large-scale applications, but it appears incremental as it adapts classical gradient descent within an existing manifold optimization framework.
The paper tackles low-rank matrix approximation by proposing a splitting numerical integration algorithm that minimizes least-squares estimation on a Riemannian manifold, with experimental results showing good scalability and satisfactory accuracy for large-scale problems.
Low rank matrix approximation is a popular topic in machine learning. In this paper, we propose a new algorithm for this topic by minimizing the least-squares estimation over the Riemannian manifold of fixed-rank matrices. The algorithm is an adaptation of classical gradient descent within the framework of optimization on manifolds. In particular, we reformulate an unconstrained optimization problem on a low-rank manifold into a differential dynamic system. We develop a splitting numerical integration method by applying a splitting integration scheme to the dynamic system. We conduct the convergence analysis of our splitting numerical integration algorithm. It can be guaranteed that the error between the recovered matrix and true result is monotonically decreasing in the Frobenius norm. Moreover, our splitting numerical integration can be adapted into matrix completion scenarios. Experimental results show that our approach has good scalability for large-scale problems with satisfactory accuracy