Advances and Challenges in Conformal Inference

We describe some recent advances in distribution-free prediction intervals in regression, using the conformal inference framework. This framework allows for the construction of a prediction band for the response variable using any estimator of the regression function. The resulting prediction band preserves the consistency properties of the original estimator under standard assumptions, while guaranteeing finite-sample marginal coverage even when these assumptions do not hold. We discuss two major variants of the conformal framework: full conformal inference and split conformal inference, along with a related jackknife method. These methods offer different tradeoffs between statistical accuracy (length of resulting prediction intervals) and computational efficiency. Time permitting, we will discuss extensions and some open challenges related to conformal inference.

This represents joint work with Rina Barber, Emmanuel Candes, Max G’Sell, Jing Lei, Aaditya Ramdas, Alessandro Rinaldo, and Larry Wasserman.

Ryan Tibshirani is an Associate Professor jointly appointed in the Departments of Statistics and Machine Learning at Carnegie Mellon University. He is the recipient of an NSF CAREER Award in 2016, and a Carnegie Mellon Teaching Innovation Award in 2017. He joined the Statistics faculty at Carnegie Mellon University in 2011, and joined the Machine Learning faculty in 2013. He did his Ph.D. in Statistics at Stanford University in 2011 (under Jonathan Taylor), and his B.S. in Mathematics at Stanford University in 2007.

]]>We describe some recent advances in distribution-free prediction intervals in regression, using the conformal inference framework. This framework allows for the construction of a prediction band for the response variable using any estimator of the regression function. The resulting prediction band preserves the consistency properties of the original estimator under standard assumptions, while guaranteeing finite-sample marginal coverage even when these assumptions do not hold. We discuss two major variants of the conformal framework: full conformal inference and split conformal inference, along with a related jackknife method. These methods offer different tradeoffs between statistical accuracy (length of resulting prediction intervals) and computational efficiency. Time permitting, we will discuss extensions and some open challenges related to conformal inference.

This represents joint work with Rina Barber, Emmanuel Candes, Max G’Sell, Jing Lei, Aaditya Ramdas, Alessandro Rinaldo, and Larry Wasserman.

]]>