Times shown are Eastern Standard Time (UTC/GMT-5)

Rackham Building,
915 E Washington St, Ann Arbor, MI 48109

In-person attendance at this event requires free advance registration. See the event home page.

Just Machine Learning

ABSTRACT

Tom Mitchell in his 1997 Machine Learning textbook defined the well-posed learning problem as follows: “A computer program is said to learn from experience E with respect to some task T and some performance measure P, if its performance on T, as measured by P, improves with experience E.” In this talk, I will discuss current tasks, experiences, and performance measures as they pertain to fairness in machine learning. The most popular task thus far has been risk assessment. For example, Jack’s risk of defaulting on a loan is 8, Jill’s is 2; Ed’s risk of recidivism is 9, Peter’s is 1. We know this task definition comes with impossibility results (e.g., see Kleinberg et al. 2016, Chouldechova 2016). I will highlight new findings in terms of these impossibility results. In addition, most human decision-makers seem to use risk estimates for efficiency purposes and not to make fairer decisions. The task of risk assessment seems to enable efficiency instead of fairness. I will present an alternative task definition whose goal is to provide more context to the human decision-maker. The problems surrounding experience have received the most attention. Joy Buolamwini (MIT Media Lab) refers to these as the “under-sampled majority” problem. The majority of the population is non-white, non-male; however, white males are overrepresented in the training data. Not being properly represented in the training data comes at a cost to the under-sampled majority when machine learning algorithms are used to aid human decision-makers. There are many well-documented incidents here; for example, facial recognition systems have poor performance on dark-skinned people. In terms of performance measures, there are a variety of definitions here from group- to individual-fairness, from anti-classification, to classification parity, to calibration. I will discuss our null model for fairness and demonstrate how to use deviations from this null model to measure favoritism and prejudice in the data.

SPEAKER BIO

Tina Eliassi-Rad is an Associate Professor of Computer Science at Northeastern University in Boston, MA. She is also a core faculty member at Northeastern University’s Network Science Institute. Prior to joining Northeastern, Tina was an Associate Professor of Computer Science at Rutgers University; and before that she was a Member of Technical Staff and Principal Investigator at Lawrence Livermore National Laboratory. Tina earned her Ph.D. in Computer Sciences (with a minor in Mathematical Statistics) at the University of Wisconsin-Madison. Her research is rooted in data mining and machine learning; and spans theory, algorithms, and applications of big data from networked representations of physical and social phenomena. She has over 80 peer-reviewed publications (including a few best paper and best paper runner-up awardees); and has given over 190 invited talks and 13 tutorials. Tina’s work has been applied to personalized search on the World-Wide Web, statistical indices of large-scale scientific simulation data, fraud detection, mobile ad targeting, cyber situational awareness, and ethics in machine learning. Her algorithms have been incorporated into systems used by the government and industry (e.g., IBM System G Graph Analytics) as well as open-source software (e.g., Stanford Network Analysis Project). In 2017, she served as the program co-chair for the ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (a.k.a. KDD, which is the premier conference on data mining) and as the program co-chair for the International Conference on Network Science (a.k.a. NetSci, which is the premier conference on network science). In 2010, she received an Outstanding Mentor Award from the Office of Science at the US Department of Energy. For more details, visit http://eliassi.org.

This event is part of the 2019 Michigan Institute for Data Science Symposium.