Wasserstein mixing time of the unadjusted Langevin algorithm
The paper claims a Wasserstein mixing-time bound of order \(\kappa \sqrt{d}/\varepsilon\) for unadjusted Langevin.
Francesco Pedrotti and Peter A. Whalley give new Wasserstein-distance estimates for the algorithm’s asymptotic bias. The setting is log-smooth, strongly log-concave measures. They say the result improves the previous state of the art by a factor of \(\sqrt{d}/\varepsilon\). ArXiv · AI/CL/LG's note
Francesco Pedrotti and Peter A. Whalley give new Wasserstein-distance estimates for the algorithm’s asymptotic bias. The setting is log-smooth, strongly log-concave measures. They say the result improves the previous state of the art by a factor of \(\sqrt{d}/\varepsilon\). ArXiv · AI/CL/LG's note
score 4