Abstract
The Schr\"odinger bridge problem (SBP) finds the most likely stochasticevolution between two probability distributions given a prior stochasticevolution. As well as applications in the natural sciences, problems of thiskind have important applications in machine learning such as dataset alignmentand hypothesis testing. Whilst the theory behind this problem is relativelymature, scalable numerical recipes to estimate the Schr\"odinger bridge remainan active area of research. We prove an equivalence between the SBP and maximumlikelihood estimation enabling direct application of successful machinelearning techniques. We propose a numerical procedure to estimate SBPs usingGaussian process and demonstrate the practical usage of our approach innumerical simulations and experiments.