A Distinguishing Attack with a Neural Network

Souza, William Augusto Rodrigues de; Tomlinson, Allan

doi:10.1109/icdmw.2013.116

Cited by 11 publications

(5 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Several previous works on using ANNs and other techniques for cipher type classification can be found [33]- [37]. However, to the best of authors' knowledge, we could not see specific previous work on using spiking ANNs for this task.…”

Section: Future Research Directionsmentioning

confidence: 86%

Intelligent Techniques in Cryptanalysis: Review and Future Directions

Al-Janabi

Al‐Khateeb

Ahmed

2017

UHDJST

View full text Add to dashboard Cite

Section: Future Research Directionsmentioning

confidence: 86%

Intelligent Techniques in Cryptanalysis: Review and Future Directions

Al-Janabi

Al‐Khateeb

Ahmed

2017

UHDJST

View full text Add to dashboard Cite

“…William et al proposed a distinction attack to identify block ciphers in [29], combining neural networks with linguistic patterns that generate signatures in ciphertexts. Employing a single 128-bit key, 240 plaintexts of 6144 and 8192 bytes in eight different languages were encrypted by the MARS, RC6, Rijndael, Serpent and Twofish algorithms.…”

Section: Crelated Researchmentioning

confidence: 99%

Post-quantum cryptographic algorithm identification using machine learning

Rocha

Xexeo

Torres

2022

J. Inf. Secur. Cryptogr.

View full text Add to dashboard Cite

This research presents a study on the identification of post-quantum cryptography algorithms through machine learning techniques. Plain text files were encoded by four post-quantum algorithms, participating in NIST's post-quantum cryptography standardization contest, in ECB mode. The resulting cryptograms were submitted to the NIST Statistical Test Suite to enable the creation of metadata files. These files provide information for six data mining algorithms to identify the cryptographic algorithm used for encryption. Identification performance was evaluated in samples of different sizes. The successful identification of each machine learning algorithm is higher than a probabilistic bid, with hit rates ranging between 73 and 100%.

show abstract

“…Methods based on statistics have gradually withdrawn from the stage of history and been replaced by machine learning methods [1]. The cryptographic algorithm recognition solution based on machine learning technology treats features as a group of attributes, equates the recognition task to the classification task, uses the training data set, which contains features and algorithm labels, to train the classifier model, and then uses the trained classifier to test Set data (unlabeled ciphertext features) for identification [2][3][4].…”

Section: Introductionmentioning

confidence: 99%

A Block Cipher Recognition Scheme Based on Deep Learning Transformer Algorithm

Yuan,

Zhang,

Zhou

et al. 2024

Preprint

View full text Add to dashboard Cite

Cryptographic algorithms are the core of data encryption, and their identification is a prerequisite for in-depth analysis of cryptography. Cryptographic algorithm identification is the process of distinguishing or identifying encryption methods by analyzing potentially distinctive information in the ciphertext when the ciphertext is known. It serves as the foundation for cryptanalysis work. The machine learning-based approach for identifying cryptographic algorithms extracts ciphertext features and trains the machine learning algorithm to build a cryptographic system identification classifier. It has the characteristics of high accuracy, concise operation process, and strong practicability. The ciphertext's complexity and the inter-data interference both rise with the number of cryptographic algorithms. This will lead to a reduction in the recognition rate of traditional machine learning cryptographic algorithm recognition solutions, poor recognition stability, and huge challenges to recognition capabilities. Deep learning algorithms have the characteristics of strong learning ability, a large amount of data processing, wide-coverage, good adaptability, and portability, etc., and have become a hot spot in the realm of cryptographic algorithm identification. In this work, a cryptographic algorithm identification methodology based on the Transformer algorithm is proposed, and the ciphertext feature extraction approach in the NIST randomness test is improved. In the research on cryptographic algorithm identification, we selected five block cipher algorithms: AES, 3DES, Blowfish, CAST, and RC2 as the research objects, and conducted two-class and five-class recognition experiments. According to experimental data, the approach suggested in this article has superior accuracy and stability than the classic machine learning model when the ciphertext size and other experimental variables are the same. Among them, on ciphertext files ranging from 1kb to 512kb, compared to the accuracy of the conventional classic machine learning algorithms SVM, GNB, KNN, RF, and LR, the average recognition accuracy of the two categories is 0.855, which is 29% higher. 33.9%, 28.5%, 34.2%, and 29.9%. Compared with the MLP accuracy of the same deep learning algorithm, the accuracy is also improved by 9%. When conducting the five-category experiment, the recognition accuracy of the scheme put forward in this piece is not less than 0.4. When the ciphertext file is 512KB, the recognition accuracy is as high as 0.42, and the average recognition accuracy of the five categories is 0.412. It is far greater than the other five classic machine learning algorithms and significantly better than the 20% random classification accuracy. Other algorithms' recognition rates fluctuate as ciphertext file sizes change. Among them, the Transformer algorithm has the smallest fluctuation and the best overall effect.

show abstract

A Distinguishing Attack with a Neural Network

Cited by 11 publications

References 5 publications

Intelligent Techniques in Cryptanalysis: Review and Future Directions

Intelligent Techniques in Cryptanalysis: Review and Future Directions

Post-quantum cryptographic algorithm identification using machine learning

A Block Cipher Recognition Scheme Based on Deep Learning Transformer Algorithm

Contact Info

Product

Resources

About