Boosted Conformal Prediction Intervals

Ran Xie,Rina Foygel Barber,Emmanuel J. Candès

DOI: https://doi.org/10.48550/arXiv.2406.07449

2024-11-10

Abstract:This paper introduces a boosted conformal procedure designed to tailor conformalized prediction intervals toward specific desired properties, such as enhanced conditional coverage or reduced interval length. We employ machine learning techniques, notably gradient boosting, to systematically improve upon a predefined conformity score function. This process is guided by carefully constructed loss functions that measure the deviation of prediction intervals from the targeted properties. The procedure operates post-training, relying solely on model predictions and without modifying the trained model (e.g., the deep network). Systematic experiments demonstrate that starting from conventional conformal methods, our boosted procedure achieves substantial improvements in reducing interval length and decreasing deviation from target conditional coverage.

Methodology,Machine Learning

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to introduce an enhanced conformal prediction intervals method to achieve specific desired properties, such as enhanced conditional coverage or reduced interval length. Specifically:

- **Problem Background**: Although traditional conformal prediction methods can ensure marginal coverage, they cannot always ensure the validity of other inferential properties (such as conditional coverage) unless additional assumptions are made. To improve these properties, researchers have proposed a variety of conformity score functions, such as Local (local adaptive score) and CQR (quantile regression conformity score).

- **Paper Objective**: This paper proposes an enhanced conformal procedure based on gradient boosting, aiming to systematically improve the predefined conformity score functions without changing the trained model. This process is guided by carefully designed loss functions that measure the deviation between the prediction intervals and the target properties. The ultimate goal is to significantly reduce the interval length and improve the conditional coverage while maintaining the marginal coverage.

In summary, the core problem of this paper is to improve the existing conformal prediction methods by introducing machine - learning techniques (especially gradient boosting) to better meet specific application requirements, especially the optimization in terms of conditional coverage and interval length.