As a point of clarification it's not the model that is being abstracted, it's the raw data. The use cases that this addresses are ones where you cannot train m

As a point of clarification it's not the model that is being abstracted, it's the raw data. The use cases that this addresses are ones where you cannot train models in the same system/space/network as where the data exists.