Learning from many trajectories