This repository contains code for "Understanding Adversarial Robustness: The Trade-off between Minimum and Average Margin".
To run the estimation:
bash estimate.sh
The script will estimate the average lower bound of the classifier in ./Model/MNISTLR/MNISTLR_01000.tar and output the result in ./Output folder.
For training and testing models, refer to train.sh and test.sh.