Timezone: »

Calibrated Data-Dependent Constraints with Exact Satisfaction Guarantees
Songkai Xue · Yuekai Sun · Mikhail Yurochkin

Tue Nov 29 09:00 AM -- 11:00 AM (PST) @ Hall J #718

We consider the task of training machine learning models with data-dependent constraints. Such constraints often arise as empirical versions of expected value constraints that enforce fairness or stability goals. We reformulate data-dependent constraints so that they are calibrated: enforcing the reformulated constraints guarantees that their expected value counterparts are satisfied with a user-prescribed probability. The resulting optimization problem is amendable to standard stochastic optimization algorithms, and we demonstrate the efficacy of our method on a fairness-sensitive classification task where we wish to guarantee the classifier's fairness (at test time).

Author Information

Songkai Xue (University of Michigan)

Education: - Ph.D. candidate, Statistics, University of Michigan, September 2020 - Present; - M.S., Applied Statistics, University of Michigan, May 2020; - B.S., Statistics, Peking University, July 2018. Research interests: - Fairness, Privacy, Robustness, and Interpretability in ML; - Learning under Distribution Shift; - Deep Learning Theory; - Statistical Network Analysis.

Yuekai Sun (University of Michigan)
Mikhail Yurochkin (IBM Research, MIT-IBM Watson AI Lab)

I am a Research Staff Member at IBM Research and MIT-IBM Watson AI Lab in Cambridge, Massachusetts. My research interests are - Algorithmic Fairness - Out-of-Distribution Generalization - Applications of Optimal Transport in Machine Learning - Model Fusion and Federated Learning Before joining IBM, I completed my PhD in Statistics at the University of Michigan, where I worked with Long Nguyen. I received my Bachelor's degree in applied mathematics and physics from Moscow Institute of Physics and Technology.

More from the Same Authors