How do you plan on getting enough data for all the training?
They reserved ecosystem bonus for data owners, also by DML protocol, data will not need to move out from the devices, so no need to compromise privacy and this makes data owners more incentives
So the data comes from a select amount of owners? How can you be sure of the quality of the data then? I'd rather see synthetic data being used on the fly, like Neuromation does. This way, there is no monopoly on the data, but rather confidence that it's of good quality, and quantity.