This code implements the Caesar, Vigenere and substitution ciphers and automatically decrypts text written in them. Katz's backoff model is used to score candidate decryptions. The Caesar cipher is can be brute forced easily, but the Vigenere and substitution ciphers use simulated annealing to find a solution.
This was a prototype to investigate the feasibility of UTEK 2018's programming challenge, hence the simplicity.
DWQI IOKDOKTO QI L ELQZJY DYBQTLJ GKO LKU IWGSJU VO OLIY DG UOTZYBD. ->
LVTUOE*WQ**J*KGB*ZIDS***Y* | THIS SENTENCE IS A FAIRLY TYPICAL ONE AND SHOULD BE EASY TO DECRYPT.
- For the Caesar and Vigenere ciphers, surprisingly little training data is needed. For the Caesar cipher, a few lines of text was enough. For the Vigenere cipher, a couple paragraphs sufficed.
- Seeding the substitution cipher with a frequency attack didn't help much.
- Optimize hyperparameters (number of iterations, speed of temperature drop)
- Try heuristic-based search: weight decrypting common letters in the ciphertext as common letters in English
- Try more organized search: swap keys from sequential indices instead of randomly
- Improve the model: use a dictionary, neural network etc.
Expects test cases in the input folder and writes to the output folder.
npm install
git clone https://github.com/utek-2018/competition-package
python3 competition-package/tester.py --ref=competition-package/