dendisuhubdy / adversarial-neural-cryptography-tensorflow

Mimicking symmetric cryptosystem using adversarial networks to show that neural nets are able to learn to communicate with each other securely

Adversarial Neural Cryptography

orginal paper : Learning to protect communications with adversarial neural cryptography

medium article link

Training parameters:

learning_rate   = 0.0008
batch_size      = 4096
sample_size     = 4096*5 # 4096 according to the paper
epochs          = 10000  # 850000 according to the paper
steps_per_epoch = int(sample_size/batch_size)


# Input and output configuration.
TEXT_SIZE = 16
KEY_SIZE  = 16

# training iterations per actors.
ITERS_PER_ACTOR = 1
EVE_MULTIPLIER = 2  # Train Eve 2x for every step of Alice/Bob

Training Loss:

First 300 epochs

Last 300 epochs