Deep machine unlearning is the problem of'removing' from a trained neural network a subset of its training set.

Machine learning models are often personalized by usinggroup attributesthat encodepersonalcharacteristics(e.g.,sex,agegroup,HIVstatus). Insuchsettings, individuals expect to receive more accurate predictions in return for disclosing group attributes to the personalized model.

Based on this analysis, our second contribution is the discovering of a type of unfairness that arises from theoretically predictable accuracy disparity across some subgroups of test nodes.