This document summarizes research on using neural networks for symmetric cryptography. It discusses how neural networks can generate secret keys through synchronization of identical tree parity machines (TPMs) that receive common random inputs. When the networks mutually learn from each other's outputs, their weight vectors synchronize to an identical state that represents the secret key. The document outlines the neural key exchange protocol and how the synchronized key can be used for encryption/decryption. It also discusses enhancing security by incorporating queries into the training process to introduce mutual influence between the communicating parties.