Do we need more 'dirty' datasets? Are we focussing too much on the same 'clean' datasets when evaluating AutoML systems?
What are the next biggest challenges to tackle AutoML?
Data augmentation. It has been used to achieve significant improvements in performance, but is pr...more
Will AutoML eliminate the need for machine learning researchers?
At some point in the future, probably.
I hope not!
Not before AGI is created
My experience at Google suggests that it is not difficult to train a programmer to be able to apply standard ML techniques. Given this, is full automation a worthwhile goal? Perhaps the focus should be on minimizing manual tuning, not eliminating it.
by Kester Tong
Maybe, we do not want Google to monopolize machine learning