Note: This Git repository uses git submodules to have access to the neural networks from literature. In order to clone them, run git submodule init and git submodule update (or copy the two git repositories in the corresponding directories if you have not cloned this repository).
| File/Directory | Explanation |
|---|---|
| cipher | Contains the implementation of all ciphers we use |
| CSYY22 | https://github.com/AI-Lab-Y/ND_mc as a submodule |
| ddt | Code for DDT calculation |
| exception | Custom exceptions |
| freshly_trained_nets | Directory where networks trained by the train_* scripts will be saved |
| Goh19 | https://github.com/agohr/deep_speck as a submodule |
| lib | External libraries |
| LLS+23 | https://github.com/JIN-smile/Improved-Related-key-Differential-based-Neural-Distinguishers as a submodule |
| nets | Previously trained networks used for most experiments in the paper |
| differential_linear_bias.py | Empirically evaluates the differential-linear biases for all one bit masks and all ciphers |
| eval.py | Provides functions for distinguisher evaluations |
| eval_(cipher).py | Evaluates the performance of the distinguishers for 'cipher' in various settings (default, real difference, etc.) |
| make_train_data.py | Provides functions for data generation |
| multi_pair_advantage.py | Evaluates the multi-pair advantage |
| present_biases.py | Statistical evaluation of the PRESENT distinguisher on more rounds |
| requirements.txt | Python packages needed to run the python scripts |
| train_(cipher).py | Trains a neural network for 'cipher' using the same hyper-parameter as in the paper |
| train_nets.py | Main code for training a neural network |
| verify_test_vectors.py | Checks that the ciphers in cipher/ are correctly implemented |
| ZWWW22 | Neural Networks from the URL provided in https://github.com/CryptAnalystDesigner/NeuralDistingsuisherWithInception/blob/main/model_file_download_address.txt |
The python code was tested using Python 3.9.12 and the packages from requirements.txt. Some experiments require up to 32GB of RAM. This can be prevented by lowering the n_samples* values, but comes at the expense of less precise results.
For experiments using the neural networks from ZWWW22, those network first need to be downloaded from the URL provided in https://github.com/CryptAnalystDesigner/NeuralDistingsuisherWithInception/blob/main/model_file_download_address.txt and then be put into the directory ZWWW22 (the directory ZWWW22 should then contain the two subdirectories Speck3264 and Simon3264).
| Identifier | Work |
|---|---|
| [CSYY22] | Chen, Y., Shen, Y., Yu, H., Yuan, S.: A New Neural Distinguisher Considering Features Derived From Multiple Ciphertext Pairs. The Computer Journal (03 2022). https://doi.org/10.1093/comjnl/bxac019, https://doi.org/10.1093/comjnl/bxac019, bxac019 |
| [Goh19] | Gohr, A.: Improving attacks on round-reduced Speck32/64 using deep learning. In: Boldyreva, A., Micciancio, D. (eds.) CRYPTO 2019, Part II. LNCS, vol. 11693, pp. 150–179. Springer, Heidelberg (Aug 2019). https://doi.org/10.1007/978-3-030-26951-7_6 |
| [LLS+23] | Lu, J., Liu, G., Sun, B., Li, C., Liu, L.: Improved (Related-Key) Differential-Based Neural Distinguishers for SIMON and SIMECK Block Ciphers. The Computer Journal (01 2023). https://doi.org/10.1093/comjnl/bxac195, https://doi.org/10.1093/comjnl/bxac195, bxac195 |
| [ZWWW22] | Zhang, L., Wang, Z., Wang, B.: Improving differential-neural cryptanalysis with inception blocks. Cryptology ePrint Archive, Report 2022/183 (2022), https://eprint.iacr.org/2022/183 |