Natural language decryption using MCMC
Using an advanced Markov Chain Monte Carlo sampling algorithm, I built an agent capable of decrypting encrypted texts.
Description
This project aimed to create an decrypter capable of iteratively decrypting text encrypted by a random mapping.
MCMC solution
The decrypter initally constructs a 2D probability grid based on bi-grams from a large corpus of English text. From here, we have the ability to give input text a score of based on how likely the text is made up of Emglish. We use MCMC sampling to allow the decrypter to explore a vast space of possible text transformations, iteratively improving the decryption process.
Usually, after around 100,000 iterations, the decrypter will reach a minimum which is hopefully the same text (or a very close approximation) of the orignal text.
