Transport for Machine Learning

Upum, transport for machine learning broadly is a very active area. A lot of work being done on both exploring the different variations of the set up. Can you comment on how computationally demanding these approaches are? I mean, it's certainly not an area that's finished. There's still a lot more to do.