ADVERSARIAL NEURAL CRYPTOGRAPHY

Based on paper Abadi, M., & Andersen, D. G. (2016). Learning to protect communications with adversarial neural cryptography. arXiv preprint arXiv:1610.06918.

Work on training neural networks for real time encryption in multi-agent transmission and multiple adversaries. Trained LSTMs, RNN's and CNN's, Alice and Bob who transmit data protecting it from multiple adversaries trying various adversarial attacks of another network Eve.

Contents

Verious Implementation of Adam, Eve, and Bob

• CNN implementations
• RNN implementations
• LSTM

Features

• Key leak
  
• Multiple advesaries
  
• Security Analysis
  
  • Chi2 and KS tests
  • Encoding