C P Mand Methology - Iterative Modeling

In c p m a i, the data is on premise and we have to do the training on premise. It's all like medical image data, and moving it to the cloud might be a regulatory headache. If your first phase does not require you going whole hog into some deep learning noral net that you have to train yourself from scratch,. then you can go back and revisit your garthem selection. And this would be one of those points of iteration to say, think big, start small, iterate.