Philip Lelyveld note: at the linked video, Microsoft demonstrates how they use predictive algorithms to move VR video processing to the cloud and create a light weight (i.e. untethered) local HMD experience with minimal latency. The potential problem, as I see it, is they quickly generate an interpolated image if their prediction is wrong. That means that, potentially, if you move your head fast, the image will snap from unfocused to focused each time the prediction is wrong.
See the original source video here:http://msrvideo.vo.msecnd.net/rmcvideos/246323/246323.mp4