Code for paper Anytime-Valid Confidence Sequences for Consistent Uncertainty Estimation in Early-Exit Neural Networks.
- Clone or download this repo.
cd
yourself to it's root directory. - Create and activate python conda enviromnent:
conda create --name eenn-avcs python=3.8
- Activate conda environment:
conda activate eenn-avcs
- Install dependencies, using
pip install -r requirements.txt
- For sythetic data experiment, see
regression_synthetic.ipynb
- For ALBERT experiment, first download precomputed logits from here and store them in
data/
folder. Then seeregression_albert.ipynb
- For MSDNet experiment, first download precomputed logits from here and store them in
data/
folder. Then seeclassification_msdnet.ipynb
The Robert Bosch GmbH is acknowledged for financial support.