config.yaml

train_classifier:
  seed: 0
  dataset: cifar-10 # {mnist, fashion-mnist, cifar-10}
  classifier: classifier-2 # {classifier-1, classifier-2}
  num_epochs: 200
  batch_size: 32
  lr: 0.01
  step_size: 10
  gamma: 0.9
  interval_log_loss: 1 # in unit epoch
  interval_checkpoint: 50 # in unit epoch

train_defensive_model:
  seed: 0
  dataset: cifar-10 # {mnist, fashion-mnist, cifar-10}
  defensive_model: defensive-model-2 # {defensive-model-1, defensive-model-2}
  num_epochs: 100
  batch_size: 256
  lr: 0.001
  step_size: 10
  gamma: 0.95
  weight_input_noise: 0.1
  weight_regularizer: 1.e-9
  interval_log_loss: 1 # in unit epoch
  interval_log_images: 10 # in unit epoch
  interval_checkpoint: 25 # in unit epoch
  num_samples: 24 # number of interval_log_images

generate_adversarial_examples:
  seed: 0
  dataset: cifar-10 # {mnist, fashion-mnist, cifar-10}
  attack_model: classifier-2 # {classifier-1, classifier-2}
  save_path: attacks

evaluate_defensive_model:
  dataset_name: cifar-10 # {mnist, fashion-mnist, cifar-10}
  defensive_models_path: models
  defensive_model_name: defensive-model-2 # {defensive-model-1, defensive-model-2}
  classifier_models_path: models
  classifier_model_name: classifier-2 # {classifier-1, classifier-2}
  attacks_data_path: attacks
  attack_model_name: classifier-2 # {classifier-1, classifier-2}
  batch_size: 1024
  result_path: results
  visualize: True
  visualization_path: visualizations