intellisec/r-admm-nonuniform

Robustness Aware Pruning ADMM with Non-Uniform Strategies

Extended Robust-ADMM(Shaokai Ye et al.) implementation with additional supports to focus on the non-unform adversarial robust pruning:

We fork the code from Robustness-Aware-Pruning-ADMM and extend with following additional supports to focus on the non-unform adversarially robust pruning:

1. Extend pruning on channel granularity in ./admm/admm.py
2. Add reading function into every configs.py in ./ADMM_examples to fetch a strategy from strategies/$dataset$.json
3. Apply non-uniform compression strategy via function insert_nonuniform_strategy in every configs.py from ./ADMM_examples.
4. Add Free-Adversarial-Training in ./ADMM_examples/imagenet/adv_main.py for ImageNet experiments.