Loading Events
  • This event has passed.
Stochastics and Statistics Seminar

Logistic Regression: The Importance of Being Improper

April 19, 2019 @ 11:00 am - 12:00 pm

Dylan Foster (MIT Institute for Foundations of Data Science)


Logistic regression is a fundamental task in machine learning and statistics. For the simple case of linear models, Hazan et al. (2014) showed that any logistic regression algorithm that estimates model weights from samples must exhibit exponential dependence on the weight magnitude. As an alternative, we explore a counterintuitive technique called improper learning, whereby one estimates a linear model by fitting a non-linear model. Past success stories for improper learning have focused on cases where it can improve computational complexity. Surprisingly, we show that for sample complexity (number of examples needed to achieve a desired accuracy level), improper learning leads to a doubly-exponential improvement in dependence on weight magnitude over estimation of model weights, and more broadly over any so-called “proper” learning algorithm. This provides a positive resolution to a COLT 2012 open problem of McMahan and Streeter. As a consequence of this improvement, we also resolve two open problems on the sample complexity of boosting and bandit multi-class classification.

Dylan Foster is a postdoctoral researcher at the MIT Institute for Foundations of Data Science. In 2018 he received his PhD in computer science at Cornell University, advised by Karthik Sridharan. His research focuses on theory for machine learning in real-world settings. He is particularly interested in all aspects of generalization theory, particularly as it applies to deep learning, non-convex optimization, and interactive learning problems including online and bandit learning. Dylan previously received his BS and MS in Electrical Engineering from USC in 2014. He has received awards including the NDSEG PhD fellowship, Facebook PhD fellowship, and best student paper award at COLT.

MIT Statistics and Data Science Center host guest lecturers from around the world in this weekly seminar.

MIT Statistics + Data Science Center
Massachusetts Institute of Technology
77 Massachusetts Avenue
Cambridge, MA 02139-4307