A-Deeper-Look-at-Machine-Learning-Based-Cryptanalysis

This is the official repository for the paper A Deeper Look at Machine Learning-Based Cryptanalysis

Requirements

This project was coded in python3.6 Requires libraries can be found in requirements.txt.

Reproductibility

To reproduce the results, you can directly run python3 run_reproduction_all.py

Section 5

To reproduce the results Table 10 and 11 of the paper, you need to run the following commands

python3 main_reproduction_speck_5round.py
python3 main_reproduction_speck_6round.py

To reproduce the results of Section 5.5, you need to run the following command

python3 main_reproduction_simon_8round.py

Section 6

To reproduce the results of Section 6, you need to run the following command

python3 main_reproduction_speck_averaging_batches.py

Demo

Colab demo:

Summary of the analysis of the pipeline

Bt running the main.py program, you can find results in the result folder that summary the run.

Files logs_lgbm.txt and res_0.npy give the results of the run.

Importance of the Masks / Features

You will find a file quality_masks.csv which compare the quality of the masks found according different criterions. There is also a study on the independance of the masks between one another in COMPARASION INTRA FEATURES XI 2.csv and with the label inINDEPENACE FEATURES LABELS.csv.

Example of feature importance for 5 round speck

This figure orders the features according their importance for the classification.

Example of decision tree for LGBM for 5 round speck

This figure shows the decision tree for the classification.

Example of how important is each bit for the decision of the neural distinguisher for 5 round speck

These figures show how important is each bit for the decision of the neural distinguisher for different methods.

Other

We also provides the possibility to change the final classifier and the inputs.

Please use this bibtex if you want to cite this repo

@misc{cryptoeprint:2021:287,
    author = {Adrien Benamira and David Gerault and Thomas Peyrin and Quan Quan Tan},
    title = {A Deeper Look at Machine Learning-Based Cryptanalysis},
    howpublished = {Cryptology ePrint Archive, Report 2021/287},
    year = {2021},
    note = {\url{https://eprint.iacr.org/2021/287}},
}

AnonymousSubmissionEuroCrypt2021/A-Deeper-Look-at-Machine-Learning-Based-Cryptanalysis