Data Decisions and Theoretical Implications when Adversarially Learning Fair Representations | Synapse