Voit lisätä twiitteihisi sijainnin, esimerkiksi kaupungin tai tarkemman paikan, verkosta ja kolmannen osapuolen sovellusten kautta. Halutessasi voit poistaa twiittisi sijaintihistorian myöhemmin. Lue lisää
elaborate?
Only disparate impact without a good reason constitutes discrimination (that's the legal definition). And in ML algorithms there's always one: optimizing the objective.
No. ML algos are in general high dimensional and non-convex. Which means that the optimizing process is path dependent and critically dependent on the architectural choices and the algorithm used. If you fail at "edge cases" (i.e. minorities), that's incompetence, not objectivity
That's true when the goal is training a model then evaluating it against a fixed test set. When models are deployed to production, two different models with the same test performance can have very different outputs for new examples of mis/under represented categories & contexts.
Various models achieve the same score on the objective but have different biases. So no, optimizing the objective is not a good reason for disparate impact.
1) „objective“ objectives don’t exist. 2) objectives = minimizing or maximizing some metric. Depending on your chosen metric-space (e.g. l1- vs l2-Norm) you decide about a sample’s relative gradient weight and thus how much your optimization will alleviate or magnify data bias.
This is illuminating with regards to your over-confidence in the incorrect perspective you keep peddling. Even with this narrow definition, it’s still also wrong in subtle ways. But at least there’s a clear line of argument.
Twitter saattaa olla ruuhkautunut tai ongelma on muuten hetkellinen. Yritä uudelleen tai käy Twitterin tilasivulla saadaksesi lisätietoja.
