Author: Matthew Gigliotti Fall 2018 A demonstration of Tree Parity Machine synchronization. This project contains an implementation of the Tree Parity Machine structure as a PyTorch module. Ideally, this implementation may be built upon to fully utilize the scalability PyTorch provides. synch.py may be altered to test the synchronization process with different parameters. In the Jupyter Notebook, we can look at plots that describe the synchronization process, and compare different learning rules and settings for K, N, and L. See https://en.wikipedia.org/wiki/Neural_cryptography for a good description of the TPM synchronization protocol. A note about this implementation: Anti-Hebbian learning appears to be disproportionately inefficient compared to Hebbian/Random Walk learning in regard to learning iterations needed for synchronization. Also, unexpected behavior has been noted for certain settings, such as k=4 and n=100. It is unclear if this can be attributed to a bug in the code. Credit to the author of https://github.com/farizrahman4u/neuralkey, who's synchronization measurements (% synchronized) I based mine off of. A list of relevant papers for those looking to learn about neural network based key exchange: https://arxiv.org/pdf/1502.05153.pdf (Neural Synchronization based Secret Key Exchange over Public Channels: A survey) http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=88E9C139C857C057C62E602F0075424D?doi=10.1.1.220.8198&rep=rep1&type=pdf (Side-Channel Resistance Evaluation of a Neural Network Based Lightweight Cryptography Scheme) https://pdfs.semanticscholar.org/d291/f30e597f1ce9ec05100cb812a93f4786e344.pdf (Neural Synchronization Using Genetic Algorithm for Secure Key Establishment) https://ieeexplore.ieee.org/document/6707125 (Authenticated key exchange protocol using neural cryptography with secret boundaries) https://www.ki.tu-berlin.de/fileadmin/fg135/publikationen/Ruttor_2006_GAN.pdf (Genetic attack on neural cryptography) http://www.cse.msu.edu/~leixinyu/Papers/TwoLayer2013.pdf (Two-layer tree-connected feed-forward neural network model for neural cryptography)