Methodology for Efficient CNN Architectures in Profiling Attacks

The current repository is associated with the article "Methodology for efficient CNN architectures in Profiling Attacks" available on the eprints

Each dataset is composed of the following scripts and repositories:

• cnn_architecture.py: provides the script in order to train the model introduced in the article,
• exploit_pred.py: computes the evolution of the right key and saves the resulted picture,
• (Optionnal) clr.py: computes the One-Cycle Policy (see "Super-Convergence: Very Fast Training of Neural Networks Using Large Learning Rates " and "A disciplined approach to neural network hyper-parameters: Part 1 -- learning rate, batch size, momentum, and weight decay,
• "training_history": contains information related to the loss function and the accuracy,
• "model_predictions": contains information related to the model predictions,
• "fig": contains the figure related to the rank evolution,
• "..._trained_models": containts the model used in the article.

The trace sets were obtained from publicly databases:

• DPA-contest v4: http://www.dpacontest.org/v4/42_traces.php
• AES_HD dataset: https://github.com/AESHD/AES_HD_Dataset
• AES_RD dataset: https://github.com/ikizhvatov/randomdelays-traces
• ASCAD: https://github.com/ANSSI-FR/ASCAD

Citation

If you use our code, models or wish to refer to our results, please use the following BibTex entry:

@misc{cryptoeprint:2019:803,
    author = {Gabriel Zaid and Lilian Bossuet and Amaury Habrard and Alexandre Venelli},
    title = {Methodology for Efficient CNN Architectures in Profiling Attacks},
    howpublished = {Cryptology ePrint Archive, Report 2019/803},
    year = {2019},
    note = {\url{https://eprint.iacr.org/2019/803}},
}