Google Brain has done something crazy again.
Using 3 roles to create a machine learning encryption and decryption system: Alice, Bob, Eve.
Alice and Bob need to communicate with each other, while Eve tries to steal data from them.
These 3 roles are neural networks.
Alice: The loss function is such that the more accurate Eve's guesses are compared to random guessing, the greater the loss.
Bob: The loss function is such that the farther the guess distance, the greater the loss.
Eve: The loss function is such that the more accurate the guess, the smaller the loss.
This forms a generative adversarial network, and the results actually allow Alice and Bob to communicate with each other, forming an encryption and decryption network, with Eve unable to obtain correct plaintext information.
This is an important example of mutual learning leading to mutual growth.
The paper in the link mentions experiments done using TensorFlow on a single GPU https://techcrunch.com/2016/10/28/googles-ai-creates-its-own-inhuman-encryption/