- This event has passed.
Lecture (data science): Algorithmic Fairness: Practical and Long-Term Challenges
May 6, 2019 @ 12:00 - 13:00
FreeAbstract
The increasing adoption of algorithmic tools has raised concerns because of their potential to perpetuate systematic subgroup discrimination when these are deployed to support decisions involving humans. Research in algorithmic fairness has studied the impact of such kinds of biases, and methods that account for subgroup fairness have been proposed. These studies have typically assessed or attacked algorithmic discrimination from the point of view of short-term target features for which these algorithms are trained, e.g. recidivism in criminal justice, hiring a job candidate, placing a kid in foster care in child welfare. Recent research identified trade-offs between accounting for fairness of these short-term outcomes and longer-term outcomes that institutions, societies and individuals usually care about. In this talk, I will first summarise the state of the art in algorithmic fairness research. Then, I will provide insights into example applications where algorithmic fairness is a challenge, how this challenge is reflected in the short and long-term, and next steps in long-term fairness research.
Biography
Diana is a final year PhD student at the School of Computer Science, The University of Auckland, New Zealand, and a Research Fellow in Data Science at the Centre for Social Data Analytics, AUT, New Zealand. Her research involves the study of learning systems that learn in the long-term, and the application of machine learning to decision making in societal problems. She is recently interested in the aspects of fairness and discrimination in the algorithm age, and how these problems are reflected in real-world decision-making processes where decisions have a long-term institutional, societal or individual impact.