Deep Learning-Assisted Differential Cryptanalysis on Round-reduced Block Ciphers: What are the advantages?

Sammanfattning

This thesis investigates and evaluates the application of deep learning techniques indifferential cryptanalysis of lightweight block ciphers. The main two focuses are oncomparison between a hybrid attack pipeline and a classical attack pipeline, and ontransfer learning compared to training a model from scratch. The study considersboth SPN ciphers and the SIMON32/64 cipher, which is part of the ISO standardfor RFID systems. Neural distinguishers based on convolutional neural networksare trained to classify pairs of ciphertexts, as either random noise or ciphertextsproduced by the cipher. For the SPN cipher it is integrated into a key recoveryattack pipeline. For SIMON32/64 it is compared to the outcome of transfer learningof a pre-trained network. The results show that a hybrid approach is comparable toa classical approach in terms of key recovery attack. The transfer learning enablesfaster convergence, but does not reach the same accuracy in classification comparedto training the model from scratch. These findings contribute to the understandingof the role of machine learning in cryptanalysis, and how it can be further studiedto potentially be more useful in the future, and what the security impacts might befor real-world use in especially supply chains using RFID technology.