Uncertainty Sets for Image Classifiers using Conformal Prediction

TL;DR

This paper introduces a simple, fast conformal prediction-based algorithm that provides formal uncertainty quantification guarantees for image classifiers, improving the size and stability of predictive sets.

Contribution

It presents a novel modification of conformal prediction that ensures finite-sample coverage guarantees and produces smaller, more stable predictive sets for image classifiers.

Findings

Achieves 5 to 10 times smaller predictive sets than baseline methods.

Provides formal coverage guarantees for any classifier and dataset.

Outperforms existing uncertainty quantification approaches on Imagenet datasets.

Abstract

Convolutional image classifiers can achieve high predictive accuracy, but quantifying their uncertainty remains an unresolved challenge, hindering their deployment in consequential settings. Existing uncertainty quantification techniques, such as Platt scaling, attempt to calibrate the network's probability estimates, but they do not have formal guarantees. We present an algorithm that modifies any classifier to output a predictive set containing the true label with a user-specified probability, such as 90%. The algorithm is simple and fast like Platt scaling, but provides a formal finite-sample coverage guarantee for every model and dataset. Our method modifies an existing conformal prediction algorithm to give more stable predictive sets by regularizing the small scores of unlikely classes after Platt scaling. In experiments on both Imagenet and Imagenet-V2 with ResNet-152 and other classifiers, our scheme outperforms existing approaches, achieving coverage with sets that are often factors of 5 to 10 smaller than a stand-alone Platt scaling baseline.