Abstract: The Schroedinger Bridge problem as generative modelling tool has been recently studied. It is known that the solution to the problem can be cast as the optimal control (Foellmer drift) which steers a diffusion with fixed marginals at times t= 0 to t= 1 such that the law of the controlled diffusion has minimal KL-divergence with respect to the Wiener measure. In those note, we review this derivation and then use the deep stochastic optimal control- a hybrid method combining deep learning and trajectory optimization- to estimate the Foellmer drift directly.