GPU-Accelerated Bregman Douglas-Rachford Splitting for Discrete Optimal Transport
The paper reports GPU BDRS implementations that beat eight GPU optimal-transport baselines across three input formats.
Xu and Ma adapt Bregman Douglas-Rachford splitting for discrete optimal transport with explicit cost matrices, point clouds, and separable grid costs. For each case, they design mathematically equivalent GPU representations aimed at stability and runtime. They benchmark all solvers on the same device and use a unified optimality measure. The authors say this is the first cross-solver GPU DOT study under that kind of comparison. ArXiv · AI/CL/LG's note
Xu and Ma adapt Bregman Douglas-Rachford splitting for discrete optimal transport with explicit cost matrices, point clouds, and separable grid costs. For each case, they design mathematically equivalent GPU representations aimed at stability and runtime. They benchmark all solvers on the same device and use a unified optimality measure. The authors say this is the first cross-solver GPU DOT study under that kind of comparison. ArXiv · AI/CL/LG's note
score 4