A space-decoupling framework for optimization on bounded-rank matrices with orthogonally invariant constraints
This work addresses a bottleneck in low-rank optimization for researchers and practitioners dealing with constrained matrix problems, though it is incremental as it builds on existing low-rank and Riemannian methods.
The paper tackles the challenge of optimizing low-rank matrices with orthogonally invariant constraints, which are hindered by coupled geometries, by proposing a space-decoupling framework that simplifies the problem into a smooth manifold for easier Riemannian algorithm implementation, and demonstrates its superiority through numerical experiments on real-world applications like spherical data fitting and deep learning.
Imposing additional constraints on low-rank optimization has garnered growing interest. However, the geometry of coupled constraints hampers the well-developed low-rank structure and makes the problem intricate. To this end, we propose a space-decoupling framework for optimization on bounded-rank matrices with orthogonally invariant constraints. The "space-decoupling" is reflected in several ways. We show that the tangent cone of coupled constraints is the intersection of tangent cones of each constraint. Moreover, we decouple the intertwined bounded-rank and orthogonally invariant constraints into two spaces, leading to optimization on a smooth manifold. Implementing Riemannian algorithms on this manifold is painless as long as the geometry of additional constraints is known. In addition, we unveil the equivalence between the reformulated problem and the original problem. Numerical experiments on real-world applications -- spherical data fitting, graph similarity measuring, low-rank SDP, model reduction of Markov processes, reinforcement learning, and deep learning -- validate the superiority of the proposed framework.