Adaptive conformal classification with noisy labels

This software repository provides a software implementation of the methods described in the following paper:

"Adaptive conformal classification with noisy labels"
Matteo Sesia, Y. X. Rachel Wang, Xin Tong
arXiv preprint https://arxiv.org/abs/2309.05092

Paper abstract

This paper develops novel conformal prediction methods for classification tasks that can automatically adapt to random label contamination in the calibration sample, leading to more informative prediction sets with stronger coverage guarantees compared to state-of-the-art approaches. This is made possible by a precise characterization of the effective coverage inflation (or deflation) suffered by standard conformal inferences in the presence of label contamination, which is then made actionable through new calibration algorithms. Our solution is flexible and can leverage different modeling assumptions about the label contamination process, while requiring no knowledge of the underlying data distribution or of the inner workings of the machine-learning classifier. The advantages of the proposed methods are demonstrated through extensive simulations and an application to object classification with the CIFAR-10H image data set.

Contents

• cln/ Python package implementing our methods
• third_party/ Third-party Python packages imported by our package.
• examples/ Jupyter notebooks with introductory usage examples
• experiments/ Code to reproduce the numerical experiments with simulated and real data discussed in the accompanying paper.

Prerequisites

Prerequisites for the cln package:

• numpy
• scipy
• sklearn
• pandas
• torch
• tqdm

Installation

The development version is available from GitHub:

git clone https://github.com/msesia/conformal-label-noise.git

License

This project is licensed under the MIT License - see the LICENSE file for details.