In this contribution, a hybrid multiscale method for simulating gas slip flow in micro-systems is proposed. Machine learning techniques are used to couple the molecular and macroscopic scale simulations.The validation steps of the learning and coupling with analytical solution are shown before presenting some physical results. This method is applied successfully and will be presented for the study of weakly rarefied flow in micro-channel of large lengths.