<ul><li>Sch"odinger bridge (SB) is a probabilistic generative model that is often applied in various domains.</li><li>A new approach, called Variational Online Mirror Descent (OMD), is proposed to enhance the stability of SB solvers.</li><li>The proposed algorithm, Variational Mirrored Schr"odinger Bridge (VMSB), utilizes the Wasserstein-Fisher-Rao geometry for Schr"odinger potentials.</li><li>Experimental results demonstrate that VMSB outperforms contemporary SB solvers and showcases the robustness predicted by the theory.</li></ul>

Variational Online Mirror Descent for Robust Learning in Schr\"odinger Bridge

Discover more