Infosec In the City (IIC)
SINCON 2021 Conference — Side-Channel Attacks at the Age of Deep-Learning — by Benjamin Timon
SINCON 2021 Conference — BizComm Talk
Day 2 (06 Nov 2021) 3.00pm—4.00pm @ Open Stage
Side-Channel Attacks at the Age of Deep-Learning
Side-Channel attacks have been around for more than 20 years and comprise multiple attack techniques exploiting devices leakages. One category of attacks called “profiled attacks” relies on machine learning algorithms to build models characterizing the leakages of devices in order to break cryptographic implementations.
As in many fields, Deep Learning has produced a paradigm shift in Side-Channel. Every year, new publications revisit existing attacks and explore brand new attacks made possible with Deep Learning.
In this talk, we review the main applications of Deep Learning for Side-Channel analysis and discuss why Deep Learning leads to exciting new possibilities in this field.
In particular, we explore the applications of the following techniques:
Convolutional Neural Networks (CNNs) to process desynchronized side-channel signals.
Data augmentation to artificially enlarge side-channel datasets.
Deep neural networks to detect leakages from implementations.
AutoEncoders and Generative Adversarial Networks (GANs) for noise reduction and training data generation.
During the talk, live demonstrations running in Python notebooks will be executed to illustrate the different techniques.
About Benjamin Timon
BENJAMIN TIMON is a senior security engineer at eShard Singapore, a start-up specialising in the security of embedded devices. He has 7+ years of experience in hardware security and side-channel attacks with past experience in security labs in the UK and Singapore. He is now part of the team developing the esDynamic software, a data science-based platform for side-channel analysis and fault injection testing. He leads the development of its AI module, where Deep Learning is used to improve the analysis of device leakages. He is also one of the developers maintaining the Scared library, and open-source framework for side-channel attacks.