Adversially Robust Sparse Models

This toolbox provides implementation for all experiments in the paper Adversarial Robustness of Supervised Sparse Coding, by J. Sulam, R. Muthukumar and R. Arora.

Experiments are separated by jupyter notebooks for clarity.

Dependencies:

• numpy 1.18.1
• torch 1.1.0
• sklearn 0.22.0
• auxiliary libs (matplotlib, pdb, time, importlib) and importantly:
• art 1.2.0: the Advesarial Robustness Toolbox

Functionality

Synthetic models and functions are gathered in adverarial_sparse_toolbox.py, whereas unsupervised and supervised dictionary models and learning functions are in sparse_learnable_dictionary.py.

---

Comments and suggestions welcome at jsulam1@jhu.edu