Events

Machine Learning and Side-Channel Analysis: Happily Ever After or Bitter Divorce?

Speaker: Aydin Aysu

Location: 370 Jay Street, Room 825

Date: Thursday, October 12, 2023

While machine learning helps automating and improving certain classification/regression tasks, side-channel analysis allows extracting secret information from systems that are mathematically secure. Although these were two fairly distinct research domains, they have recently been getting closer. Yet it is unclear how they would couple. In this talk, I will explain my research group's efforts in using machine learning for side channels, and side channels for machine learning. I will first demonstrate new side-channel attacks on machine-learning hardware that can steal valuable machine-learning models and methods for protection. Then, I will show the use of novel machine-learning techniques to outperform classical side-channel attacks and break cryptographic systems that cannot be broken with classical attacks.