The initial implementation of the HENN: Hyper Evidential Deep Learning to Quantify Composite Classification Uncertainty which was accepted in ICLR 2024.

config.yml contains all experimental settings for our HENN model.

The folder "data" contains all data preparation files for different datasets, including tinyImageNet, Living17, Nonliving26, CIFAR100, and NAbirds.

Each data file contains different processing methods for different methods: HENN, DNN, and ENN, etc.

• GDD_main.py This is the main file to run our HENN code.

• GDD_train.py This file contains the model training detail.

• GDD_evaluate.py This file contains the evaluation of the HENN model in validation set.

• GDD_test.py This file contains the evaluation metric calculation during test phase of the HENN model.

• baseline_DetNN.py is the main file for baseline DNN.
• baseline_ENN.py is the main file for baseline ENN.

• helper_functions.py contains the necessary data preparation functions and necessary functions for HENN training.

• All results are saved in cloud using wandb.
• roc_draw_plot.ipynb For figures we generated, such as AUROC curves.

If you find HENN useful in your work, please consider citing the following BibTeX entry:

@inproceedings{
li2024hyper,
title={Hyper Evidential Deep Learning to Quantify Composite Classification Uncertainty},
author={Changbin Li and Kangshuo Li and Yuzhe Ou and Lance M. Kaplan and Audun J{\o}sang and Jin-Hee Cho and DONG HYUN JEONG and Feng Chen},
booktitle={The Twelfth International Conference on Learning Representations},
year={2024},
url={https://openreview.net/forum?id=A7t7z6g6tM}
}