adversarial_attacks_and_sparse_representations

This repo allows to recreate the experiments and plots from the paper "On the Relationship Between Universal Adversarial Attacks and Sparse Representations".

Recreate the figures

All the run files are configured with valid defaults and documented with the available flagged changes.

Fig. 1, 6:

./evaluate_DA_on_classification.py

Fig. 2:

./scatter_plot_attack_delta_sc.py

Fig. 3, 4:

./load_corrs_and_plot.py

Fig. 5:

./hist_delta_energy.py

Fig. 7:

evaluate_DA_on_classification.py

Train sparse coding models

To train a sparse coder, please run

./train_sparse_coding_models/train_sparse_coder.py

To train a linear classifier over the sparse code, please run

./train_sparse_coding_models/train_linear_classifier_over_sparse_code.py

Don't forget to cite!

If you find this repo useful or mention us in your work, please don't forget to cite us.

@article{weitzner2023relationship, title={On the Relationship Between Universal Adversarial Attacks and Sparse Representations}, author={Weitzner, Dana and Giryes, Raja}, journal={IEEE Open Journal of Signal Processing}, volume={4}, pages={99--107}, year={2023}, publisher={IEEE} }