Mix-n-Match-Calibration

This repository contains code that accompanies the paper Mix-n-Match: Ensemble and Compositional Methods for Uncertainty Calibration in Deep Learning. Please see the paper for more details.

LLNL CP Number: CP02333

Citation

If you find this library useful please consider citing our paper:

@inproceedings{zhang2020mix,
  author={Zhang, Jize and Kailkhura, Bhavya and Han, T},
  booktitle={International Conference on Machine Learning (ICML)},
  title = {Mix-n-Match: Ensemble and Compositional Methods for Uncertainty Calibration in Deep Learning},
  year = {2020},
}

To use in a project

The file demo_calibration.py is a template to conduct calibration and evaluate their performance with various methods.

The file util_calibration.py contains the functions describing the proposed mix-n-match calibration methods.

The file util_evaluation.py contains the functions describing the proposed mix-n-match evaluation methods.