Constructing a Fair Classifier with Generated Fair Data

Fairness in machine learning is getting rising attention as it directly related to real-world applications and social problems. Recent methods have been explored alleviate the discrimination between certain demographic groups that are characterized by sensitive attributes (such race, age, or gender). Some studies found data itself biased, so training on causes unfair decision making. Models trained raw can replicate even exacerbate bias prediction groups. This leads vastly different performance In order address this issue, we propose a new approach improve fairness generating fair data. We introduce generative model generate cross-domain samples w.r.t. multiple attributes. ensures infinite number of balanced \wrt both target label enhance prediction. By classifier solely with synthetic then transfer real data, overcome under-representation problem which non-trivial since collecting extremely time resource consuming. provide empirical evidence demonstrate benefit our respect accuracy.