High dimensional online calibration in polynomial time
This resolves open questions in sequential forecasting for high-dimensional settings, offering a polynomial-time solution to a previously exponential-time problem.
The paper tackles the curse of dimensionality in online calibration by presenting the first asymptotically calibrated strategy that achieves non-trivial calibration in polynomial time, specifically requiring T = d^{O(1/ε^2)} days for accuracy ε, and proves a lower bound of T = d^{Ω(log(1/ε))} rounds.
In online (sequential) calibration, a forecaster predicts probability distributions over a finite outcome space $[d]$ over a sequence of $T$ days, with the goal of being calibrated. While asymptotically calibrated strategies are known to exist, they suffer from the curse of dimensionality: the best known algorithms require $\exp(d)$ days to achieve non-trivial calibration. In this work, we present the first asymptotically calibrated strategy that guarantees non-trivial calibration after a polynomial number of rounds. Specifically, for any desired accuracy $ε> 0$, our forecaster becomes $ε$-calibrated after $T = d^{O(1/ε^2)}$ days. We complement this result with a lower bound, proving that at least $T = d^{Ω(\log(1/ε))}$ rounds are necessary to achieve $ε$-calibration. Our results resolve the open questions posed by [Abernethy-Mannor'11, Hazan-Kakade'12]. Our algorithm is inspired by recent breakthroughs in swap regret minimization [Peng-Rubinstein'24, Dagan et al.'24]. Despite its strong theoretical guarantees, the approach is remarkably simple and intuitive: it randomly selects among a set of sub-forecasters, each of which predicts the empirical outcome frequency over recent time windows.