Machine learning for cryptanalysis
Lightweight cryptography involves cryptography subject to constraints such as area and power consumption. In 2019, NIST organised a currently ongoing competition for lightweight authenticated encryption. This competition, currently in its final stage, has narrowed down the initial 57 submissions...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
Nanyang Technological University
2022
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/156805 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | Lightweight cryptography involves cryptography subject to constraints such as
area and power consumption. In 2019, NIST organised a currently ongoing
competition for lightweight authenticated encryption. This competition, currently
in its final stage, has narrowed down the initial 57 submissions to 10
finalists, of which includes the cipher known as ASCON. Concurrent with this
was a research paper published by Gohr in 2019. Here, it was shown that it
was possible to apply deep learning to cryptanalysis. More specifically, it was
possible to design a neural distinguisher for the Speck 32/64 cipher, where for
a specified input difference, it was possible to distinguish between ciphertext
pairs that had this input difference and a random one via classification.
In this study, we provide an amalgamation of these two advances. Here, we
first attempt to apply Gohr’s neural network architecture to the round function
of the ASCON cipher. Following this, we attempt to improve on the architecture
in hopes of greater accuracy. The primary result found was that Gohr’s network
architecture had the ability to learn well up to 3.5 rounds of the ASCON round
function. Following this, we were able to provide modifications for marginal
improvement in accuracy of the 4 round case.
Keywords - cryptanalysis, deep learning, lightweight cryptography |
---|