Parallelly Sliced Optimal Transport on Spheres and on the Rotation Group

Michael Quellmalz,Léo Buecher,Gabriele Steidl
2024-05-22
Abstract:Sliced optimal transport, which is basically a Radon transform followed by one-dimensional optimal transport, became popular in various applications due to its efficient computation. In this paper, we deal with sliced optimal transport on the sphere $\mathbb{S}^{d-1}$ and on the rotation group SO(3). We propose a parallel slicing procedure of the sphere which requires again only optimal transforms on the line. We analyze the properties of the corresponding parallelly sliced optimal transport, which provides in particular a rotationally invariant metric on the spherical probability measures. For SO(3), we introduce a new two-dimensional Radon transform and develop its singular value decomposition. Based on this, we propose a sliced optimal transport on SO(3).
Numerical Analysis,Optimization and Control
What problem does this paper attempt to address?
### What problems does this paper attempt to solve? This paper aims to solve the Optimal Transport (OT) problems on the sphere \( S^{d - 1} \) and the three - dimensional rotation group \( SO(3) \). Specifically, the author proposes the Parallelly Sliced Optimal Transport (PSOT) method to simplify and accelerate the optimal transport calculations on these geometric spaces. #### Main research contents: 1. **Parallel Sliced Optimal Transport on the Sphere**: - The author proposes a new parallel slicing method, which transforms the high - dimensional transport problem on the sphere \( S^{d - 1} \) into a one - dimensional line segment transport problem. - This method provides a rotation - invariant metric and is 40 to 100 times faster in calculation than the existing semicircle slicing method. - Through this new method, the author derives the parallel - sliced Wasserstein distance and proves its relationship with the standard Wasserstein distance. 2. **Sliced Optimal Transport on the Three - Dimensional Rotation Group \( SO(3) \)**: - The author introduces a new two - dimensional Radon transform and develops its Singular Value Decomposition (SVD), laying the foundation for sliced optimal transport on \( SO(3) \). - Based on this new Radon transform, the author proposes a sliced optimal transport method on \( SO(3) \) and proves that the corresponding sliced Wasserstein distance satisfies the metric properties on the set of probability measures. 3. **Barycenter Calculation**: - The author studies the barycenter calculation method based on the new sliced Wasserstein distance, including algorithms in the free - support and fixed - support discrete settings. - Provides numerical experiment results, showing the performance of these algorithms on the two - dimensional sphere, and compares them with the existing semicircle slicing method and the regularized Wasserstein barycenter. #### Formula Representation: - The parallel slicing operator is defined as: \[ S_{\psi}: S^{d - 1} \to I, \quad S_{\psi}(\xi) := \langle \xi, \psi \rangle \] where \( \psi \in S^{d - 1} \), \( I = [-1, 1] \). - The parallel - sliced Wasserstein distance is defined as: \[ \text{PSW}_p^p(\mu, \nu) := \int_{S^{d - 1}} W_p^p((S_{\psi})_{\#}\mu, (S_{\psi})_{\#}\nu) \, d u_{S^{d - 1}}(\psi) \] - On \( SO(3) \), the slicing operator is defined as: \[ d_Q: SO(3) \to [0, \pi], \quad P \mapsto \angle(Q^T P) \] where \( \angle(Q) := \arccos\left(\frac{\text{trace}(Q) - 1}{2}\right) \). - The sliced Wasserstein distance on \( SO(3) \) is defined as: \[ \text{SOSW}_p^p(\mu, \nu) := \int_{SO(3)} W_p^p((d_Q)_{\#}\mu, (d_Q)_{\#}\nu) \, d u_{SO(3)}(Q) \] Through these methods, the author not only solves the optimal transport problems on the sphere and the rotation group, but also improves the computational efficiency and provides theoretical guarantees.