Séminaire sécurité des systèmes électroniques embarqués

Accueil     Présentation     Archives

Eleonora Cagli


Machine Learning Approach for Profiling Side-Channel Attacks

Side-Channel Attacks aim at recognizing the manipulation of a cryptographic secret variable, by observing noisy signals coming from a physical device performing the cryptographic computation. A Facebook tool aims at recognizing the faces of people invited at your last party, by observing a possibly out of focus picture you posted last night. There is a strict analogy between these two tasks. The high success rate achieved in the second outlined context might justify a Side-Channel attacker trying to exploit some similar techniques.
Nowadays in image recognition and many other domains, the most successful techniques are based on Machine Learning, and in particular on Convolutional Neural Networks. In this talk we will see how an attacker can take advantage of the Machine Learning approach. Indeed, it allows him to construct a model of his side-channel signals, integrating in the automatic learning some signal preprocessing phases that represent today a great limitation to the profiling attacks optimality.