i.e. how to quantify a business goal as a scalar that can be readily computed from a model's output. Requires extensive domain knowledge
Once we solve AutoML, the automated creation of machine learning pipelines, we can move on to focus on the art of loss function engineering
-
-
-
Also requires experimentation - often only a successful A/B test tells if your loss function is appropriate
End of conversation
New conversation -
-
-
Can you make smartphone alarms that read out the words I've entered instead of playing chime music? "Ding a ling" v "dinner ready in 10"
Thanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
-
I believe automatic generation of models is what he means ...
End of conversation
New conversation -
-
-
And will be helpful to develop reactive and realtime-feedback-providing UIs like
@machine_ui to have the system be more surveyableThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
Researchers will spend time defining the problem they want to solve.
-
So instead of randomly picking new goals, and then solving them, they will randomly pick goals and they will be solved for them?
- Show replies
New conversation -
-
-
I think in many areas the loss function is easier (and in others it's insanely hard). Does automl include feature engineering?
-
I think people are limiting themselves to poor problem specifications (e.g. image classification vs. "vision") bc the ML part is hard enough
- Show replies
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.