Audrey Chang is a Finalist in Harvard College's Three Minute Thesis Competition

Senior Joint Concentrator in Statistics and Computer Science Audrey Chang was recently selected as a finalist for Harvard College's Three Minute Thesis competition.  The Three Minute Thesis competition was developed originally by The University of Queensland, Australia.  At Harvard, seniors submit an accessible and engaging presentation geared towards a broad audience for the first round of the contest and then present their project to a live audience in the second round.  

Chang's thesis is on the topic of classification in statistics.  In classification, recent work has shown that models produce overconfident predictions for atypical inputs — inputs that are rare, unusual, or underrepresented for their type. This poses a fairness concern because atypical inputs are more likely to belong to societally underrepresented groups. Chang generalizes the concept of atypicality from classification to regression to explicitly address this overconfidence, resulting in predicted confidence that is similarly reliable across low-atypicality and high-atypicality inputs.