Log-Averaged Mirror Prox for Fast, Large-Scale Optimal Transport in Linear Space
For practitioners of large-scale optimal transport, LAMP provides a memory-efficient primal-dual method that scales to previously intractable problem sizes.
LAMP reduces storage complexity of primal-dual optimal transport from O(nm) to O(n+m) while maintaining competitive arithmetic complexity, enabling scaling to problems with 2^18 marginal supports, and outperforming first-order baselines in high-accuracy settings.
We propose Log-Averaged Mirror Prox (LAMP), a linear-space primal-dual method for large-scale optimal transport. LAMP implements primal mirror prox updates by tracking an averaged dual sequence, reducing storage complexity from ${O}(nm)$ to $O(n+m)$ while preserving dense, GPU-friendly reductions. Consequently, LAMP preserves the last-iterate $\widetilde{O}( nm\varepsilon^{-1})$ arithmetic complexity of conservatively parameterized primal-dual mirror prox. We further analyze LAMP as a direct optimal transport solver in a more performant parameter regime, providing a last-iterate sub-optimality certificate dependent on infeasibility and an explicit $O(1/t)$ term. Moreover, we give a computable sufficient condition for best-iterate convergence to a saddle-point. Numerical experiments with an optimized CUDA implementation show that LAMP outperforms first-order baselines in several high-accuracy (entropic) optimal transport problems. LAMP is further shown to scale up to problems with $n=m=2^{18}$ marginal supports, which were previously beyond the reach of primal-dual first-order methods.