A1018
Title: Persuasive calibration
Authors: Wei Tang - Chinese University of Hong Kong (Hong Kong) [presenting]
Yiding Feng - Hong Kong University of Science and Technology (Hong Kong)
Abstract: The persuasive calibration problem is studied, where a principal aims to provide trustworthy predictions about underlying events to a downstream agent to make desired decisions. The standard calibration framework is adopted that regulates predictions to be unbiased, conditional on their own value, and thus, they can reliably be interpreted at the face value by the agent. Allowing a small calibration error budget, the aim is to answer the following question: What is and how the optimal predictor can be computed under this calibration error budget, especially when there exists incentive misalignment between the principal and the agent? A general framework is developed by viewing predictors as post-processed versions of perfectly calibrated predictors. The structure of the optimal predictor is characterized. On the algorithmic side, an FPTAS is provided to compute the approximately optimal predictors for a general setting. Moreover, for the L1- and L-Infinity-norm ECE, polynomial-time algorithms are provided that compute the exact optimal predictor.
