vrt1shjwlkr/ndss21-model-poisoning

Code for NDSS 2021 Paper "Manipulating the Byzantine: Optimizing Model Poisoning Attacks and Defenses Against Federated Learning"

ndss21-model-poisoning's Issues

Realization of PGA

Dear author, where can I find the code of the PGA attack shown in the paper?

Fang's adaptive attack to Krum

In compute_lambda_fang function, Code for Fang attack on krum, Cell 7,

term_1 = min_score / ((n_benign - n_attackers - 1) * torch.sqrt(torch.Tensor([d]))[0])

should be
term_1 = min_score / ((n_benign - 2 * n_attackers - 1) * torch.sqrt(torch.Tensor([d]))[0])
based on equation 3 from the original paper

Please correct me if I am wrong :)

The realization of the Dnc in paper.

Can I have the relevant code for the defense mechanism, please?

Fang's adaptive attack to Median and Trmean.

In get_malicious_updates_fang_trmean function, `Fang's adaptive attack to Median and Trmean , Cell 7,

mal_vec = (torch.stack([(deviation > 0).type(torch.FloatTensor)] * max_rand.shape[1]).T.cuda() * max_rand + torch.stack( [(deviation > 0).type(torch.FloatTensor)] * min_rand.shape[1]).T.cuda() * min_rand).T .

Should it be
mal_vec = (torch.stack([(deviation < 0).type(torch.FloatTensor)] * max_rand.shape[1]).T.cuda() * max_rand + torch.stack( [(deviation > 0).type(torch.FloatTensor)] * min_rand.shape[1]).T.cuda() * min_rand).T instead? As it should sample from [w_max, b * w_max] if deviation < 0.