More specifically, imagine we were examining a machine learning system for bias in a certain dimension and for simplicity sake it consists of a singular machine learning algorithm. Could we take that model before it was trained, make one copy of the original data set and introduce artificial bias in one direction(of the dimension in question), make another data set with artificial bias in the other direction, train the algorithm on each data set, and then compare results across all three models? Or asked in another way, could we create some type of lower and upper bound to give us a probability of bias? Thanks for any answers or resources on the matter!


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.