An algorithm for calibrating forecasts of high-dimensional outcomes.

We propose a novel calibration measure, referred to as (pseudo) KL-Calibration, and leverage it to establish several new bounds on the swap regret for a range of significant loss functions.