Kilroy Was Here: The First Step Towards Explainability of Neural Networks in Profiled Side-Channel Analysis

Daan Valk; Stjepan Picek; Shivam Bhasin

doi:10.1007/978-3-030-68773-1_9

Kilroy Was Here: The First Step Towards Explainability of Neural Networks in Profiled Side-Channel Analysis

Daan Valk, Stjepan Picek^*, Shivam Bhasin

^*Corresponding author for this work

Cyber Security

Research output: Chapter in Book/Conference proceedings/Edited volume › Conference contribution › Scientific › peer-review

3 Citations (Scopus)

Abstract

While several works have explored the application of deep learning for efficient profiled side-channel analysis, explainability, or, in other words, what neural networks learn remains a rather untouched topic. As a first step, this paper explores the Singular Vector Canonical Correlation Analysis (SVCCA) tool to interpret what neural networks learn while training on different side-channel datasets, by concentrating on deep layers of the network. Information from SVCCA can help, to an extent, with several practical problems in a profiled side-channel analysis like portability issue and criteria to choose a number of layers/neurons to fight portability, provide insight on the correct size of training dataset and detect deceptive conditions like over-specialization of networks.

Original language	English
Title of host publication	Constructive Side-Channel Analysis and Secure Design - 11th International Workshop, COSADE 2020, Revised Selected Papers
Editors	Guido Marco Bertoni, Francesco Regazzoni
Publisher	Springer
Pages	175-199
Number of pages	25
Volume	12244
ISBN (Print)	9783030687724
DOIs	https://doi.org/10.1007/978-3-030-68773-1_9
Publication status	Published - 2021
Event	11th International Workshop on Constructive Side-Channel Analysis and Secure Design, COSADE 2020 - Lugano, Switzerland Duration: 1 Apr 2020 → 3 Apr 2020

Publication series

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume	12244 LNCS
ISSN (Print)	0302-9743
ISSN (Electronic)	1611-3349

Conference

Conference	11th International Workshop on Constructive Side-Channel Analysis and Secure Design, COSADE 2020
Country/Territory	Switzerland
City	Lugano
Period	1/04/20 → 3/04/20

Keywords

Deep learning
Neural networks
Representation learning
Side-channel analysis

Access to Document

10.1007/978-3-030-68773-1_9

Cite this

Valk, D., Picek, S., & Bhasin, S. (2021). Kilroy Was Here: The First Step Towards Explainability of Neural Networks in Profiled Side-Channel Analysis. In G. M. Bertoni, & F. Regazzoni (Eds.), Constructive Side-Channel Analysis and Secure Design - 11th International Workshop, COSADE 2020, Revised Selected Papers (Vol. 12244, pp. 175-199). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12244 LNCS). Springer. https://doi.org/10.1007/978-3-030-68773-1_9

Valk, Daan ; Picek, Stjepan ; Bhasin, Shivam. / Kilroy Was Here : The First Step Towards Explainability of Neural Networks in Profiled Side-Channel Analysis. Constructive Side-Channel Analysis and Secure Design - 11th International Workshop, COSADE 2020, Revised Selected Papers. editor / Guido Marco Bertoni ; Francesco Regazzoni. Vol. 12244 Springer, 2021. pp. 175-199 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{3340d8e7208b4508b8fd06c998a8c384,

title = "Kilroy Was Here: The First Step Towards Explainability of Neural Networks in Profiled Side-Channel Analysis",

abstract = "While several works have explored the application of deep learning for efficient profiled side-channel analysis, explainability, or, in other words, what neural networks learn remains a rather untouched topic. As a first step, this paper explores the Singular Vector Canonical Correlation Analysis (SVCCA) tool to interpret what neural networks learn while training on different side-channel datasets, by concentrating on deep layers of the network. Information from SVCCA can help, to an extent, with several practical problems in a profiled side-channel analysis like portability issue and criteria to choose a number of layers/neurons to fight portability, provide insight on the correct size of training dataset and detect deceptive conditions like over-specialization of networks.",

keywords = "Deep learning, Neural networks, Representation learning, Side-channel analysis",

author = "Daan Valk and Stjepan Picek and Shivam Bhasin",

year = "2021",

doi = "10.1007/978-3-030-68773-1_9",

language = "English",

isbn = "9783030687724",

volume = "12244",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer",

pages = "175--199",

editor = "Bertoni, {Guido Marco} and Francesco Regazzoni",

booktitle = "Constructive Side-Channel Analysis and Secure Design - 11th International Workshop, COSADE 2020, Revised Selected Papers",

note = "11th International Workshop on Constructive Side-Channel Analysis and Secure Design, COSADE 2020 ; Conference date: 01-04-2020 Through 03-04-2020",

}

Valk, D, Picek, S & Bhasin, S 2021, Kilroy Was Here: The First Step Towards Explainability of Neural Networks in Profiled Side-Channel Analysis. in GM Bertoni & F Regazzoni (eds), Constructive Side-Channel Analysis and Secure Design - 11th International Workshop, COSADE 2020, Revised Selected Papers. vol. 12244, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 12244 LNCS, Springer, pp. 175-199, 11th International Workshop on Constructive Side-Channel Analysis and Secure Design, COSADE 2020, Lugano, Switzerland, 1/04/20. https://doi.org/10.1007/978-3-030-68773-1_9

Kilroy Was Here: The First Step Towards Explainability of Neural Networks in Profiled Side-Channel Analysis. / Valk, Daan; Picek, Stjepan; Bhasin, Shivam.
Constructive Side-Channel Analysis and Secure Design - 11th International Workshop, COSADE 2020, Revised Selected Papers. ed. / Guido Marco Bertoni; Francesco Regazzoni. Vol. 12244 Springer, 2021. p. 175-199 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 12244 LNCS).

Research output: Chapter in Book/Conference proceedings/Edited volume › Conference contribution › Scientific › peer-review

TY - GEN

T1 - Kilroy Was Here

T2 - 11th International Workshop on Constructive Side-Channel Analysis and Secure Design, COSADE 2020

AU - Valk, Daan

AU - Picek, Stjepan

AU - Bhasin, Shivam

PY - 2021

Y1 - 2021

N2 - While several works have explored the application of deep learning for efficient profiled side-channel analysis, explainability, or, in other words, what neural networks learn remains a rather untouched topic. As a first step, this paper explores the Singular Vector Canonical Correlation Analysis (SVCCA) tool to interpret what neural networks learn while training on different side-channel datasets, by concentrating on deep layers of the network. Information from SVCCA can help, to an extent, with several practical problems in a profiled side-channel analysis like portability issue and criteria to choose a number of layers/neurons to fight portability, provide insight on the correct size of training dataset and detect deceptive conditions like over-specialization of networks.

AB - While several works have explored the application of deep learning for efficient profiled side-channel analysis, explainability, or, in other words, what neural networks learn remains a rather untouched topic. As a first step, this paper explores the Singular Vector Canonical Correlation Analysis (SVCCA) tool to interpret what neural networks learn while training on different side-channel datasets, by concentrating on deep layers of the network. Information from SVCCA can help, to an extent, with several practical problems in a profiled side-channel analysis like portability issue and criteria to choose a number of layers/neurons to fight portability, provide insight on the correct size of training dataset and detect deceptive conditions like over-specialization of networks.

KW - Deep learning

KW - Neural networks

KW - Representation learning

KW - Side-channel analysis

UR - http://www.scopus.com/inward/record.url?scp=85102313268&partnerID=8YFLogxK

U2 - 10.1007/978-3-030-68773-1_9

DO - 10.1007/978-3-030-68773-1_9

M3 - Conference contribution

AN - SCOPUS:85102313268

SN - 9783030687724

VL - 12244

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 175

EP - 199

BT - Constructive Side-Channel Analysis and Secure Design - 11th International Workshop, COSADE 2020, Revised Selected Papers

A2 - Bertoni, Guido Marco

A2 - Regazzoni, Francesco

PB - Springer

Y2 - 1 April 2020 through 3 April 2020

ER -

Valk D, Picek S, Bhasin S. Kilroy Was Here: The First Step Towards Explainability of Neural Networks in Profiled Side-Channel Analysis. In Bertoni GM, Regazzoni F, editors, Constructive Side-Channel Analysis and Secure Design - 11th International Workshop, COSADE 2020, Revised Selected Papers. Vol. 12244. Springer. 2021. p. 175-199. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-68773-1_9

Kilroy Was Here: The First Step Towards Explainability of Neural Networks in Profiled Side-Channel Analysis

Abstract

Publication series

Conference

Keywords

Access to Document

Other files and links

Fingerprint

Cite this