Rescheduled from 12/2 to 12/16 (same time and location).
Discrimination as Data Perturbation: Supervised Learning Framework for Prevention and Evaluation
Abstract: The actions of individuals can be discriminatory with respect to certain protected attributes, such as race or gender. Recently, discrimination has become a focal concern in artificial decision-making systems. These systems are trained with potentially discriminatory data and might inadvertently introduce biases against certain groups. An important question is how to train computational models without inducing discrimination. We contribute to this growing area of research by i) defining direct and indirect discrimination as respective perturbations of data; ii) introducing a novel evaluation framework to assess the performance of supervised learning methods for discrimination prevention; and iii) proposing a novel discrimination prevention method that performs better than several state-of-the-art methods in the proposed evaluation framework.
Bio: Professor Grabowicz joined the College of Information and Computer Sciences of the University of Massachusetts Amherst in Fall 2018 as a Research Assistant Professor. Prior to this, from 2013 to 2018, he was a postdoctoral researcher at the Max Planck Institute for Software Systems in Saarbruecken, Germany. He received Ph.D. in Interdisciplinary Physics from University of Balearic Islands (2013) and M.Sc. in Applied Physics at Warsaw University of Technology (2008). Professor Grabowicz's interdisciplinary research contributes statistical methods to understand and augment social processes in social computing systems.