Adversarial-Neural-Network-Simulation-using-GANs

We are trying to see if neural networks can learn to use secret keys to protect communications from other neural networks. Specifically, we are focusing on ensuring the confidentiality in a network with multiple agents and we specify these properties in the form of an adversary. Thus, we are trying to limit the information gleaned by Eve from a communication between Bob and Alice by eavesdropping on them. There are no specific cryptographic algorithms being assigned to these networks, rather we train them end to end adversarially.