FastSinkhorn delivers a stable CUDA log-domain Sinkhorn solver achieving 12x speedup over POT and 5.9x over PyTorch baselines on n=m=8192 problems while using only 256 MB GPU memory.
Mathematics of Computation , volume =
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Fast Log-Domain Sinkhorn Optimal Transport with Warp-Level GPU Reductions
FastSinkhorn delivers a stable CUDA log-domain Sinkhorn solver achieving 12x speedup over POT and 5.9x over PyTorch baselines on n=m=8192 problems while using only 256 MB GPU memory.