Computational and Statistical Hardness of Calibration Distance

TL;DR

This paper investigates the computational complexity of calculating and estimating the calibration distance, a key measure of miscalibration, providing efficient algorithms under certain conditions and proving NP-hardness when assumptions are relaxed.

Contribution

It introduces an efficient exact algorithm for uniform, noiseless cases, extends it to a polynomial-time approximation scheme, and establishes sample complexity bounds for estimation, along with new hardness proofs.

Findings

Efficient exact computation for uniform, noiseless distributions.

NP-hardness when assumptions are relaxed.

Sample complexity of Θ(1/ε^3) for estimation.

Abstract

The distance from calibration, introduced by B{\l}asiok, Gopalan, Hu, and Nakkiran (STOC 2023), has recently emerged as a central measure of miscalibration for probabilistic predictors. We study the fundamental problems of computing and estimating this quantity, given either an exact description of the data distribution or only sample access to it. We give an efficient algorithm that exactly computes the calibration distance when the distribution has a uniform marginal and noiseless labels, which improves the $O(1/\sqrt{|\mathcal{X}|})$ additive approximation of Qiao and Zheng (COLT 2024) for this special case. Perhaps surprisingly, the problem becomes $\mathsf{NP}$-hard when either of the two assumptions is removed. We extend our algorithm to a polynomial-time approximation scheme for the general case. For the estimation problem, we show that $\Theta(1/\epsilon^3)$ samples are sufficient and necessary for the empirical calibration distance to be upper bounded by the true distance plus $\epsilon$. In contrast, a polynomial dependence on the domain size -- incurred by the learning-based baseline -- is unavoidable for two-sided estimation. Our positive results are based on simple sparsifications of both the distribution and the target predictor, which significantly reduce the search space for computation and lead to stronger concentration for the estimation problem. To prove the hardness results, we introduce new techniques for certifying lower bounds on the calibration distance -- a problem that is hard in general due to its $\textsf{co-NP}$-completeness.