Learning from Discriminatory Training Data. (arXiv:1912.08189v4 [cs.LG] UPDATED)

Supervised learning systems are trained using historical data and, if the
data was tainted by discrimination, they may unintentionally learn to
discriminate against protected groups. We propose that fair learning methods,
despite training on potentially discriminatory datasets, shall perform well on
fair test datasets. Such dataset shifts crystallize application scenarios for
specific fair learning methods. For instance, the removal of direct
discrimination can be represented as a particular dataset shift problem. For
this scenario, we propose a learning method that …

application arxiv data dataset datasets discrimination error fair historical data learn shift supervised learning systems test test datasets training training data