The Role of Data Management in Studying Fairness

Fairness is a heavily studied topic in Machine Learning. The purpose of this talk is to survey prominent work on fairness and argue for the need for a data management approach. I will illustrate that with a recent study we did in the context of TaskRabbit and Google job search, two online job marketplaces. A data management framework would enable a variety of queries such as finding the k least/most discriminated against groups/jobs/locations. The large number of groups, queries and locations calls for using efficient algorithms. I will end the talk with new opportunities that merit our attention.