Cryptanalysis

Citation Author(s):
Bhuvaneshwari
A J
Sri Sivasubramaniya Nadar College of Eng
Kaythry
P
Sri Sivasubramaniya Nadar College of Eng
Submitted by:
Bhuvaneshwari A J
Last updated:
Wed, 03/05/2025 - 03:28
DOI:
10.21227/hs9r-aj87
Data Format:
*.txt
Links:
Deep-Learning-Based Cryptanalysis of Lightweight Block Ciphers Revisited
License:
Creative Commons Attribution
7 Views
Categories:
Artificial Intelligence
Communications
Security
Keywords:
cryptanalysis, Lightweight Cryptography, Ascon
0
0 ratings - Please login to submit your rating.
ACCESS DATASET CITE SHARE/EMBED

Abstract 

This paper explores the cryptanalysis of the ASCON algorithm, a lightweight cryptographic method designed for applications like the Internet of Things (IoT). We utilize deep learning techniques to identify potential vulnerabilities within ASCON's structure. First, we provide an overview of how ASCON operates, including key generation and encryption processes. We generate a dataset of encrypted samples and apply various deep learning models, such as Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) networks, to detect weaknesses in the encryption.A detailed bitwise accuracy analysis revealed that the models' predictions for each bit hovered around 50%, indicating a performance level akin to random guessing. Specifically, Bit 1 and Bit 7 achieved slightly above 50.03% accuracy, while Bit 4 reached 50.04%. The remaining bits scored below 50%, confirming the challenges in effectively breaking ASCON's encryption. This research highlights the effectiveness of ASCON in resisting cryptanalysis and the limitations faced by machine learning methods in this context.

Instructions: 

To analyze ciphertext and plaintext pairs for weaknesses in the ASCON algorithm, we propose a deep learning (DL) methodology that leverages sophisticated architecture tailored for cryptanalysis. Our dataset consists of 800,000 training samples and 200,000 testing samples generated using the ASCON cipher. The preprocessing phase involves converting both input (plaintext and ciphertext) and output (encryption keys) data into binary format, ensuring uniformity for deep learning processing. The model is trained to predict the encryption keys through a structured approach using a hybrid ResNet architecture with 25 residual blocks, enhanced by a Gated Linear Unit (GLU) layer for optimized feature extraction. During training, we employ the Adam optimizer to efficiently minimize the binary cross-entropy loss over 1000 epochs, with a batch size of 64. To ensure the robustness of the model and mitigate overfitting, we meticulously monitor validation loss and accuracy throughout the training process. The final model aims to accurately predict 8-bit keys from the provided data pairs, testing the resilience of the ASCON algorithm against cryptanalytic attacks

More from this Author

Cryptanalysis

Dataset Files

    Files have not been uploaded for this dataset

    Documentation

    AttachmentSize
    File Dataset for cryptanalysis of ASCON readme file.docx17.33 KB

     

    QUESTIONS?