Thinning for Accelerating the Learning of Point Processes