Deep Leakage From Gradients [arXiv]

Launch multi-run based on Hydra Sweeper:

python main.py -m --config-name=experiment

Launch single-run:

python main.py -m <options>

model parameters

[ model ] = { McMahan_CNN, Custom_ResNet}

[ model.init_method ] = { xavier_normal_, xavier_uniform_ , kaiming_normal_, kaiming_uniform_ }

[ model.activation ] = {ReLU, Sigmoid}

---

Attacker Parameters

[ attacker.optimizer.type ] = { Adam, LBFGS, SGD }

[ attacker.optimizer.learning_rate ] = any float

[ attacker.optimizer.scheduler ] = { True, False }

[ attacker.loss_function ] = { MSE, cosine_similarity }

---

Client Parameters

[ client.prune.type ] = { random, small, None }

[ client.prune.percentage ] = any float between 0-1