Cyber Security Webinar by Dr. Guilherme Perin : Deep Learning Generalizations for Profiled Side-channel Analysis
16 June 2020 12:00 till 12:45 - Location: ZOOM-MEETING - DETAILS WILL BE ANNOUNCED
Recently, deep neural networks (DNN) have demonstrated impressive performance in the profiled side-channel analysis (SCA) context. However, the training of these complex learning systems still faces a lot of difficulties, especially related to the choice of numerous hyper-parameters.
Besides that, deep learning metrics are not very consistent for the SCA context, which can easily lead to misleading conclusions about leakage detection and overfitting. In this presentation, we will propose solutions for both cases: (1) how to improve the performance of DNN when several hyper-parameters needs to be tested for SCA and (2) how to find the correct metrics to reduce overfitting during the training phase.
In particular, we present solutions based on the usage of ensembles and information bottleneck theory as an optimal metric for SCA.
The presentation will take 30mins and 15 minutes Q+A time.