Source code for the paper: Surrogate approach to uncertainty quantification of neural networks for regression
- data/* - data files
- model/* - trained model files
- regression_model.py - training and inference functions for the given regression network
- surrogates.py - functions for the proposed surrogates: Input perturbation, Gradient norm, MC-dropout, Knowledge distillation, Ensemble
- uncertainty_score.py - functions for calculating the uncertainty score from the proposed surrogate
- main.py - script for overall running code
- data_stats.py - summary statistics of datasets
- utils.py
- Python
- PyTorch
- NumPy
- scikit-learn