Distinguished Lecture Series
2018-2019 Speakers

Machine Learning as a Tool for Scientific Discovery: Applications to Behavioral Economics and Medicine

Thursday, September 20, 2018

Abstract:
I will describe efforts I have been involved in to use machine learning as a way to push the frontier of scientific understanding. The goal in these projects is to both test existing theories or to generate new ones. In testing hypotheses, these algorithms prove particularly useful because they allow us to go past testing the accuracy of a theory to quantifying its completeness. I illustrate using data on human biases in understanding randomness - a topic of long-standing interest in behavioral science. We find that existing models explain only a small fraction of the explainable variance. These algorithms also prove helpful in generating new hypothesis because they can detect signal the human mind and existing theoriesignore. I illustrate with rich EKG data that we model using a deep learning pipeline. The resulting algorithm manages to find unexpected signal in the EKGs for social and physiological outcomes---signal that is not predicted by existing theories. These discoveries raises an interpretability issue - what is the algorithm seeing and what new mechanisms does it suggest? In this context, interpretability takes on a slight twist. The goal is not to make the algorithm palatable to human intuitions - or to build trust. Instead the goal is to find the part of the algorithm that is counter-intuitive, so as to generate new hypotheses. We develop a technique ("anchored representations') that makes black box algorithms interpretable in this way. When applied to our EKG findings and shown to clinicians, intriguing hypotheses arise. As a whole, this work has not just suggested not just the utility of algorithms but raised new conceptual questions.

Bio:
Sendhil Mullainathan is a University Professor and Professor of Computation and Behavioral Science at Chicago Booth. He has worked on machine learning, poverty, finance behavioral economics and a wide variety of topics such as: the impact of poverty on mental bandwidth; whether CEO pay is excessive; using fictitious resumes to measure discrimination; showing that higher cigarette taxes makes smokers happier; modeling how competition affects media bias; and a model of coarse thinking. His latest research focuses on using machine learning to better understand human behavior.

Preventing Fairness Gerrymandering in Machine Learning

Monday, January 14, 2019

Abstract:
The most prevalent notions of fairness in machine learning are statistical definitions: they fix a small collection of pre-defined attributes (such as race, gender, age or disability), and then ask for parity of some statistic of the classifier across these attributes. Constraints of this form are susceptible to intentional or inadvertent "fairness gerrymandering", in which a classifier appears to be fair on each attribute marginally, but badly violates the fairness constraint on one or more structured subgroups (such as disabled Hispanic women over age 55). We instead propose statistical notions of fairness binding across exponentially(or infinitely) many subgroups, defined by a structured class of functions over the protected attributes. This interpolates between statistical definitions of fairness and recently proposed individual notions of fairness, but raises several interesting computational challenges.

We describe an algorithm that provably converges to the best subgroup-fair classifier. This algorithm is based on a formulation of subgroup fairness as a two-player zero-sum game between a Learner and an Auditor. We provide an extensive empirical evaluation of our algorithm on a number of fairness-sensitive datasets.

Joint work with Seth Neel, Aaron Roth and Zhiwei Steven Wu, based on papers at ICML 2018 and FAT* 2019.

See, Hear, Move: Towards Embodied Visual Perception

Thursday, January 24, 2019

Abstract:
Computer vision has seen major success in learning to recognize objects from massive “disembodied” Web photo collections labeled by human annotators. Yet cognitive science tells us that perception develops in the context of acting and moving in the world---and without intensive supervision. Meanwhile, many realistic vision tasks require not only categorizing a well-composed human-taken photo, but also actively deciding where to look in the first place and how to interact with the environment. In the context of these challenges, we are exploring ways to learn visual representations from unlabeled video accompanied by multi-modal sensory data like egomotion and sound. Moving from passively captured video to agents that control their own first-person cameras, we investigate how agents can learn to move intelligently to acquire visual observations. Finally, we explore how to learn object affordances from video of complex human behavior; we show that anticipating a new object’s function actually improves the agent’s ability to recognize it.

Bio:
Kristen Grauman is a Professor in the Department of Computer Science at the University of Texas at Austin and a Research Scientist at Facebook AI Research. Her research in computer vision and machine learning focuses on visual recognition and search. Before joining UT Austin in 2007, she received her Ph.D. at MIT in computer science. She is a Sloan Fellow, a recipient of NSF CAREER and ONR Young Investigator awards, the 2013 PAMI Young Researcher Award, the 2013 IJCAI Computers and Thought Award, a Presidential Early Career Award for Scientists and Engineers (PECASE), a 2017 Helmholtz Prize computer vision “test of time” award, and the 2018 J.K. Aggarwal Prize from the International Association for Pattern Recognition. She and her collaborators were recognized with best paper awards at CVPR 2008, ICCV 2011, and ACCV 2016. She previously served as Program Chair of the Conference on Computer Vision and Pattern Recognition (CVPR) in 2015 and currently serves as Associate Editor-in-Chief for the journal Pattern Analysis and Machine Intelligence (PAMI) and as a Program Chair of Neural Information Processing Systems (NeurIPS) 2018.