This is the link to the research paper ("Learning to protect communications with adversarial neural cryptography"): https://arxiv.org/abs/1610.06918