Join us at TensorFlow World, Oct 28-31. Use code TF20 for 20% off select passes. Register now

nsl.lib.get_target_indices

View source on GitHub

Selects targeting classes for adversarial attack (classification only).

nsl.lib.get_target_indices(
    logits,
    labels,
    adv_target_config
)

Args:

  • logits: tensor of shape=[batch_size, num_classes], dtype=tf.float32.
  • labels: int tensor for ground truth, shape=[batch_size].
  • adv_target_config: contains types of adversarial targeting.

Returns:

Tensor of shape [batch_size] and dtype=tf.int32 of indices of targets.