Testing Group Fairness via Optimal Transport Projections

              Nian Si1 Karthyek Murthy 2 Jose Blanchet1 Viet Anh Nguyen1 3

             Abstract                A natural first explanation for the reported algorithmic bi-
   We present a statistical testing framework to de-     ases is that the data used to train the algorithms may al-
   tect if a given machine learning classifier fails to    ready be corrupted by human biases (Buolamwini & Ge-
   satisfy a wide range of group fairnes ...