The Perils of Learning from Biased Data

Georgia Tech's Judy Hoffman: The Perils of Learning from Biased Data

April 23, 2021

Judy HoffmanGeorgia Tech University Assistant Professor Judy Hoffman presented The Perils of Learning from Biased Data at the Applications of Data Science and AI to Equity, Race, and Inclusion in Mobility and Transportation Seminar on April 23, 2021, co-sponsored by ITS Berkeley, the College of Engineering, and CITRIS. This topic brings a unique and innovative perspective to existing discussions around diversity, equity, and inclusion. Our aim with these series is to reflect on and raise awareness of applications, opportunities, and potential misuses of these techniques in the mobility and transportation space, specifically as it refers to race, equity, and diversity.

Abstract: A key task for safely deploying autonomous vehicles is to have reliable perception and understanding of the visual world. Modern computer vision systems can leverage large manually annotated visual datasets to learn visual recognition models that can seemingly replicate the behavior of the human annotators, automatically recognizing cars on the road and pedestrians crossing the street. In this talk, I will discuss a central challenge with learning-based systems which is that they rely on sufficiently diverse training data to capture all data variance anticipated at deployment. When the data used for learning in fact only represents a biased subset of the world, the model may suffer poor predictive performance when the test time bias changes. I will explore how this situation arises under benign conditions, such as weather pattern changes, as well as how systematic demographic bias can lead to inequitable predictive performance across subpopulations.

Dr. Judy Hoffman is an Assistant Professor in the School of Interactive Computing at Georgia Tech and a member of the Machine Learning Center. Her research lies at the intersection of computer vision and machine learning with specialization in domain adaptation, transfer learning, adversarial robustness, and algorithmic fairness. She has been awarded the NVIDIA female leader in computer vision award in 2020, AIMiner top 100 most influential scholars in Machine Learning (2020), MIT EECS Rising Star in 2015, and is a recipient of the NSF Graduate Fellowship. In addition to her research, she co-founded and continues to advise for Women in Computer Vision, an organization which provides mentorship and travel support for early-career women in the computer vision community. Prior to joining Georgia Tech, she was a Research Scientist at Facebook AI Research. She received her PhD in Electrical Engineering and Computer Science from UC Berkeley in 2016 after which she completed Postdocs at Stanford University (2017) and UC Berkeley (2018).