SLIDE 1
Learning when to stop
Ileana Buhan @ileanabuhan
2
Learning when to stop Ileana Buhan @ileanabuhan Our story.. 2020 - - PowerPoint PPT Presentation
Learning when to stop Ileana Buhan @ileanabuhan Our story.. 2020 - - PowerPoint PPT Presentation
Learning when to stop Ileana Buhan @ileanabuhan Our story.. 2020 2017 2005 1997 1941 2 Its 1941 3 Its 1997 4 Its 2005 5 SCA (non-profiled) AES(k) closed sample 6 SCA (non-profiled) AES(k) AES
SLIDE 2
SLIDE 3
3
Its 1941
SLIDE 4
4
Its 1997…
SLIDE 5
5
Its 2005…
SLIDE 6
6
SCA (non-profiled)
AES(k) closed sample
𝑛
SLIDE 7
7
SCA (non-profiled)
AES(k) AES
Leakage model Distinguisher
𝑛 𝑛 𝑙
SLIDE 8
8
SCA (non-profiled)
AES(k)
AES
𝑙!
AES AES
𝑛! 𝑛" 𝑛#
AES
𝑙"
AES AES
𝑛! 𝑛" 𝑛#
AES
𝑙$
AES AES
𝑛! 𝑛" 𝑛#
AES
𝑙%
AES AES
𝑛! 𝑛" 𝑛#
𝑒(𝑙!) 𝑒(𝑙") 𝑒(𝑙#) 𝑒(𝑙$)
𝑛!, 𝑛", 𝑛$. . 𝑛# 𝑙& ≈ 𝑙
SLIDE 9
11
SCA (profiled)
AES
𝑛 𝑙
- pen sample
SLIDE 10
14
Its 2017
SLIDE 11
15
AES
𝑛
AES(k)
𝑙
- pen sample
closed sample
𝑛
learning ATTACK
SCA (profiled)
SLIDE 12
16
TRACES BUILD the MACHINE USE the MACHINE
Acquisition Learning/Profiling Attack
Workflow
TRAIN
SLIDE 13
17
TRACES BUILD the MACHINE USE the MACHINE
Acquisition Learning/Profiling Attack
Workflow
TRAIN CONSTRUCT TRAIN
SLIDE 14
18
But..
SLIDE 15
19
TRACES BUILD the MACHINE USE the MACHINE
Acquisition Learning/Profiling Attack
Workflow
TRAIN CONSTRUCT TRAIN
SLIDE 16
20
But..
Perin G., Buhan I.R., Picek S., Learning when to stop:a mutual information approach to fight overfitting in profiled side-channel analysis a mutual information approach to fight overfitting in profiled side-channel analysis (Submitted to CHES 2020);
SLIDE 17
Leakage characterisation using deep networks
Perin G., Buhan I.R., Picek S., Learning when to stop:a mutual information approach to fight overfitting in profiled side-channel analysis a mutual information approach to fight overfitting in profiled side-channel analysis (Submitted to CHES 2020);
data labels Input Layer Hidden Layers Output Layer
𝑌, 𝑍
SLIDE 18
Leakage characterisation using deep networks
lets give a name to this layer Input Layer Hidden Layers Output Layer
𝑌, 𝑍 𝑈
!
SLIDE 19
Leakage characterisation using deep networks
Perin G., Buhan I.R., Picek S., Learning when to stop:a mutual information approach to fight overfitting in profiled side-channel analysis a mutual information approach to fight overfitting in profiled side-channel analysis (Submitted to CHES 2020);
Input Layer Hidden Layers Output Layer the information this layer has about the data
𝐽(𝑌; 𝑈
!), 𝐽(𝑈 !; 𝑍)
𝑈
!
𝑌, 𝑍
SLIDE 20
Leakage characterisation using deep networks
Perin G., Buhan I.R., Picek S., Learning when to stop:a mutual information approach to fight overfitting in profiled side-channel analysis a mutual information approach to fight overfitting in profiled side-channel analysis (Submitted to CHES 2020);
Input Layer Hidden Layers Output Layer the information layer T1 has about the labels
𝐽(𝑌; 𝑈
!), 𝐽(𝑈 !; 𝑍)
𝑈
!
𝑌, 𝑍
SLIDE 21
Leakage characterisation using deep networks
Perin G., Buhan I.R., Picek S., Learning when to stop:a mutual information approach to fight overfitting in profiled side-channel analysis a mutual information approach to fight overfitting in profiled side-channel analysis (Submitted to CHES 2020);
Input Layer Hidden Layers Output Layer
𝐽(𝑌; 𝑈"), 𝐽(𝑈"; 𝑍)
𝑈
!
𝑌, 𝑍 𝑈"
SLIDE 22
Leakage characterisation using deep networks
Perin G., Buhan I.R., Picek S., Learning when to stop:a mutual information approach to fight overfitting in profiled side-channel analysis a mutual information approach to fight overfitting in profiled side-channel analysis (Submitted to CHES 2020);
Input Layer Hidden Layers Output Layer
𝐽(𝑌; 𝑈
#), 𝐽(𝑈 #; 𝑍)
𝑈
!
𝑌, 𝑍 𝑈" 𝑈
#
SLIDE 23
Leakage characterisation using deep networks
Perin G., Buhan I.R., Picek S., Learning when to stop:a mutual information approach to fight overfitting in profiled side-channel analysis a mutual information approach to fight overfitting in profiled side-channel analysis (Submitted to CHES 2020);
Input Layer Hidden Layers Output Layer
𝐽(𝑌; 𝑇), 𝐽(𝑇; 𝑍)
𝑈
!
𝑌, 𝑍 𝑈" 𝑈
#
𝑇
SLIDE 24
How a network learns
Perin G., Buhan I.R., Picek S., Learning when to stop:a mutual information approach to fight overfitting in profiled side-channel analysis a mutual information approach to fight overfitting in profiled side-channel analysis (Submitted to CHES 2020);
SLIDE 25
How the network learns
validation set
Best epoch
SLIDE 26
Leakage characterisation using deep networks
Perin G., Buhan I.R., Picek S., Learning when to stop:a mutual information approach to fight overfitting in profiled side-channel analysis a mutual information approach to fight overfitting in profiled side-channel analysis (Submitted to CHES 2020);
GE AND Success Rate on ASCAD database
SLIDE 27
Leakage characterisation using deep networks
Perin G., Buhan I.R., Picek S., Learning when to stop:a mutual information approach to fight overfitting in profiled side-channel analysis a mutual information approach to fight overfitting in profiled side-channel analysis (Submitted to CHES 2020);
GE AND Success Rate on CHES AES database
SLIDE 28
Message to my younger self
never stop believing in yourself never stop being the your best version help Jane with security evaluations
SLIDE 29
SLIDE 30
2003 Numerical Methods Full-time teacher 2002 Mathematics and Computer Science Oct 2004 PhD, University of Twente Oct 2008 2008 (Senior) Research Scientist Philips Research 2011 Product Manager Training 2010 consultant Siemens Guest Researcher Radboud University 2019