RC-EENN

This is the public code repository for our paper: Fast yet Safe: Early-Exiting with Risk Control

Setup

1. Clone or download this repo. cd yourself to it's root directory.
2. Create and activate python conda enviromnent: conda create --name rc-eenn python=3.10
3. Activate conda environment: conda activate rc-eenn
4. Install dependencies, using pip install -r requirements.txt

TODO: add requirements for dee_diff and sem_seg experiments

Code

Code for each experiment can be found in its respective subfolder:

• Image classification (ImageNet) --> img_cls
• Semantic segmentation (Cityscapes, GTA5) --> sem_seg
• Language modeling (SQuAD, CNN/DM) --> calm
• Image generation with early-exit diffusion (CelebA, CIFAR) --> dee_diff

Acknowledgements

The Robert Bosch GmbH is acknowledged for financial support.

License

TODO

Citation

If you find this repository helpful, please consider citing:

@article{jazbec2024fast,
    title = {Fast yet Safe: Early-Exiting with Risk Control}, 
    author = {Metod Jazbec and Alexander Timans and Tin Hadži Veljković and Kaspar Sakmann and Dan Zhang and Christian A. Naesseth and Eric Nalisnick},
    journal = {Arxiv Preprint},
    year = {2024},
}