Fair Classication with Noisy Protected Attributes:
                A Framework with Provable Guarantees

             L. Elisa Celis 1 Lingxiao Huang 2 Vijay Keswani 1 Nisheeth K. Vishnoi 3

              Abstract                ξ 0 := (f =1). Then, given a performance function, fairness
    We present an optimization framework for learn-       constraints in classication impose a feasible classier to
    ing a fair classier in the presence of noisy per-      have similar performance for all ...