Low-Rank Optimal Transport through Factor Relaxation with Latent Coupling
About
Optimal transport (OT) is a general framework for finding a minimum-cost transport plan, or coupling, between probability distributions, and has many applications in machine learning. A key challenge in applying OT to massive datasets is the quadratic scaling of the coupling matrix with the size of the dataset. [Forrow et al. 2019] introduced a factored coupling for the k-Wasserstein barycenter problem, which [Scetbon et al. 2021] adapted to solve the primal low-rank OT problem. We derive an alternative parameterization of the low-rank problem based on the $\textit{latent coupling}$ (LC) factorization previously introduced by [Lin et al. 2021] generalizing [Forrow et al. 2019]. The LC factorization has multiple advantages for low-rank OT including decoupling the problem into three OT problems and greater flexibility and interpretability. We leverage these advantages to derive a new algorithm $\textit{Factor Relaxation with Latent Coupling}$ (FRLC), which uses $\textit{coordinate}$ mirror descent to compute the LC factorization. FRLC handles multiple OT objectives (Wasserstein, Gromov-Wasserstein, Fused Gromov-Wasserstein), and marginal constraints (balanced, unbalanced, and semi-relaxed) with linear space complexity. We provide theoretical results on FRLC, and demonstrate superior performance on diverse applications -- including graph clustering and spatial transcriptomics -- while demonstrating its interpretability.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Wasserstein Distance Estimation | Fragmented hypercube | Absolute Error (W₂²)0.526 | 40 | |
| Low-rank Optimal Transport | CIFAR-10 | OT Cost235.9 | 3 | |
| Low-rank Optimal Transport | Mouse embryo E8.5 -> E8.75 | OT Cost0.553 | 3 | |
| Single-cell transcriptomics alignment | Single-cell mouse embryo transcriptomics E8.5 → E8.75 | OT Cost0.553 | 3 | |
| Single-cell transcriptomics alignment | Single-cell mouse embryo transcriptomics E8.75 → E9.0 | OT Cost0.405 | 3 | |
| Low-rank Optimal Transport | Mouse embryo E9.5 -> E9.75 | OT Cost0.399 | 2 | |
| Single-cell transcriptomics alignment | Single-cell mouse embryo transcriptomics E9.0 → E9.25 | OT Cost0.481 | 2 | |
| Single-cell transcriptomics alignment | Single-cell mouse embryo transcriptomics E9.25 → E9.5 | OT Cost0.431 | 2 | |
| Single-cell transcriptomics alignment | Single-cell mouse embryo transcriptomics E9.5 → E9.75 | OT Cost0.399 | 2 | |
| Single-cell transcriptomics alignment | Single-cell mouse embryo transcriptomics E9.75 → E10.0 | OT Cost0.379 | 2 |