Neural Information Processing Systems http://nips.cc/

Machine learning has provided a universal toolbox enhancing the decision making in many disciplines from advertising and recommender systems toeducation andcriminal justice.

The impossibility theorem of fairness is a foundational result in the algorithmic fairness literature. It states that outside of special cases, one cannot exactly and simultaneously satisfy all three common and intuitive definitions of fairness demographic parity, equalized odds, and predictive rate parity. This result has driven most works to focus on solutions for one or two of the metrics.

Figure 1 shows the change of accuracy under different cutoff value. However, for gender classification under CelebA dataset, thetrade-offbetweenλval and accuracyisnotveryclear;and wesuspect that under suchscenario, focusing on hard samples does not harm the performance of easy samples, and thus benefits the classifier. Figure 1 shows the change of fairness (equalized odds) under different cutoff value. Suppose we have a large unlabeled training set of sizeN and a small labeled validation set { xvalj,yvalj,1 j M} with M N. In each training step, we sample a small mini-batch of size n(n < N) from training set and perform random augmentation twice to obtain a subset { xi,1 i 2n} and we update the contrastive encoderf with parameterθ. During validation, we freeze the contrastive encoder and train a downstream linear classifierg with parameterω for classification task.

Next, we apply our proposed fitting method (see Section 2) on this data set.

As the use of machine learning models in real world high-stakes decision settings continues to grow, it is highly important that we are able to audit and control for any potential fairness violations these models may exhibit towards certain groups. To do so, one naturally requires access to sensitive attributes, such as demographics, biological sex, or other potentially sensitive features that determine group membership. Unfortunately, in many settings, this information is often unavailable. In this work we study the well known equalized odds (EOD) definition of fairness. In a setting without sensitive attributes, we first provide tight and computable upper bounds for the EOD violation of a predictor. These bounds precisely reflect the worst possible EOD violation. Second, we demonstrate how one can provably control the worst-case EOD by a new post-processing correction method. Our results characterize when directly controlling for EOD with respect to the predicted sensitive attributes is -- and when is not -- optimal when it comes to controlling worst-case EOD. Our results hold under assumptions that are milder than previous works, and we illustrate these results with experiments on synthetic and real datasets.