pith. sign in

arxiv: 1907.02674 · v1 · pith:5VPE5HKTnew · submitted 2019-07-05 · 📡 eess.SP

Practical Approaches Towards Deep-Learning Based Cross-Device Power Side Channel Attack

Anupam Golder , Debayan Das , Josef Danial , Santosh Ghosh , Shreyas Sen , Arijit Raychowdhury This is my paper

Pith reviewed 2026-05-25 02:21 UTC · model grok-4.3

classification 📡 eess.SP
keywords power side-channel attackcross-device attackdeep learningAES-128principal component analysismulti-layer perceptrondynamic time warpingAVR microcontroller
0
0 comments X

The pith

PCA pre-processing and multi-device training let an MLP recover AES-128 key bytes at 99.43 percent accuracy across 30 AVR devices despite hardware variations.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper shows that profiling-based cross-device power side-channel attacks on AES-128 become practical when deep learning is combined with targeted pre-processing. Significant device-to-device differences in power traces from 8-bit AVR microcontrollers normally degrade attack success. Applying Principal Component Analysis before feeding traces into a multi-layer perceptron trained on data from several devices yields an average 99.43 percent accuracy on the first key byte for all thirty devices in the set. Adding Dynamic Time Warping to align misaligned traces further improves results, and the MLP-PCA pipeline beats a CNN baseline by roughly twenty percent on aligned data and more than ten percent on misaligned data.

Core claim

Utilizing Principal Component Analysis based pre-processing and multi-device training, a Multi-Layer Perceptron based 256-class classifier can achieve an average accuracy of 99.43 percent in recovering the first key byte from all the 30 devices even in the presence of significant inter-device variations; DTW combined with PCA followed by the same MLP further raises accuracy by at least 10.97 percent over CNN approaches for traces with up to 50 sample misalignments.

What carries the argument

PCA-based pre-processing that reduces trace dimensionality while preserving key-dependent features, paired with training a 256-class MLP classifier on power traces collected from multiple devices.

If this is right

  • The MLP with PCA pre-processing outperforms a CNN trained on four devices by about 20 percent in average cross-device test accuracy on aligned traces.
  • DTW alignment followed by PCA and the 256-class MLP maintains at least 10.97 percent higher accuracy than CNN methods even when traces differ by up to 50 time samples.
  • Cross-device key recovery reaches 99.43 percent average accuracy on a set of 30 devices using only the first key byte of AES-128.
  • Multi-device training plus dimensionality reduction makes profiling attacks viable despite hardware manufacturing differences.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same pre-processing steps could be tested on other microcontrollers or ciphers to check whether the accuracy gain generalizes beyond 8-bit AVR AES implementations.
  • Collecting traces from even more devices during training might further reduce the accuracy gap between seen and unseen hardware.
  • If PCA discards too little key information, the method might combine with other dimensionality techniques such as autoencoders for additional robustness.

Load-bearing premise

Device-to-device variations in power traces are large enough to break single-device models yet remain correctable by PCA and multi-device training without erasing the key-dependent information needed for classification.

What would settle it

A measured drop in first-byte recovery accuracy below usable levels when the same MLP-PCA pipeline is tested on a fresh set of AVR devices never seen during training or on traces with misalignments exceeding 50 samples.

Figures

Figures reproduced from arXiv: 1907.02674 by Anupam Golder, Arijit Raychowdhury, Debayan Das, Josef Danial, Santosh Ghosh, Shreyas Sen.

Figure 1
Figure 1. Figure 1: Box-and-whisker plots showing a distribution of 180 observations of sample #96 of the power traces obtained from [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: ChipWhisperer platform for capturing and recording power traces from an AVR XMega microcontroller running AES [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Overview of Dynamic Time Warping (DTW): Warp path for traces X and Y shows how both the traces are non-linearly [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Architecture of the proposed Multi-Layer Perceptron for Cross-device Side Channel Attack. The input layer consists of [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Test Accuracy of MLP classifier when number of devices in training set is (a) one (b) two (c) three (d) four. As can be [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Rationale behind Device 18 being an outlier in Figure 5(a): [PITH_FULL_IMAGE:figures/full_fig_p009_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Boundary of Bivariate Probability Density Function (PDF) for different scenarios. Two of the most prominent leakage [PITH_FULL_IMAGE:figures/full_fig_p009_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Architecture of 1-D CNN: Samples of raw traces are directly fed to the input layer. Two 1-D convolutional layers [PITH_FULL_IMAGE:figures/full_fig_p010_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Test Accuracy of CNN classifier when number of devices in training set is (a) one (b) four. As can be seen from (a), [PITH_FULL_IMAGE:figures/full_fig_p011_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Principal Component Analysis of Raw Traces: (a) Transformed Trace after PCA (first 1000 Time Samples). Note that, [PITH_FULL_IMAGE:figures/full_fig_p012_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Performance of PCA-MLP: (a) Average Test Accuracy(%) vs. number of principal components used (for Training [PITH_FULL_IMAGE:figures/full_fig_p012_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Ten Traces superimposed on each other (first 200 Samples shown) when they are (a) misaligned randomly up to 50 [PITH_FULL_IMAGE:figures/full_fig_p013_12.png] view at source ↗
read the original abstract

Power side-channel analysis (SCA) has been of immense interest to most embedded designers to evaluate the physical security of the system. This work presents profiling-based cross-device power SCA attacks using deep learning techniques on 8-bit AVR microcontroller devices running AES-128. Firstly, we show the practical issues that arise in these profiling-based cross-device attacks due to significant device-to-device variations. Secondly, we show that utilizing Principal Component Analysis (PCA) based pre-processing and multi-device training, a Multi-Layer Perceptron (MLP) based 256-class classifier can achieve an average accuracy of 99.43% in recovering the first key byte from all the 30 devices in our data set, even in the presence of significant inter-device variations. Results show that the designed MLP with PCA-based pre-processing outperforms a Convolutional Neural Network (CNN) with 4-device training by ~20%in terms of the average test accuracy of cross-device attack for the aligned traces captured using the ChipWhisperer hardware.Finally, to extend the practicality of these cross-device attacks, another pre-processing step, namely, Dynamic Time Warping (DTW) has been utilized to remove any misalignment among the traces, before performing PCA. DTW along with PCA followed by the 256-class MLP classifier provides >=10.97% higher accuracy than the CNN based approach for cross-device attack even in the presence of up to 50 time-sample misalignments between the traces.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper claims that profiling-based cross-device power side-channel attacks on AES-128 running on 8-bit AVR devices can be made practical using deep learning. It shows that PCA-based pre-processing combined with multi-device training enables an MLP 256-class classifier to recover the first key byte with 99.43% average accuracy across all 30 devices despite inter-device variations. The MLP+PCA approach outperforms a 4-device CNN baseline by approximately 20% on aligned traces, and adding DTW pre-processing maintains at least 10.97% higher accuracy than CNN even with up to 50-sample misalignments.

Significance. If the reported accuracies are reproducible, the work demonstrates a concrete, practical mitigation of device-to-device variation in SCA via standard pre-processing (PCA, DTW) and multi-device profiling, which could inform security evaluations of embedded cryptographic implementations. The empirical outperformance over a CNN baseline on real ChipWhisperer traces provides a useful data point for the community.

major comments (3)
  1. [Abstract] Abstract: The 99.43% average accuracy figure is reported without any accompanying information on the total number of traces per device, the train/test split (how many of the 30 devices are used for training versus testing), or statistical measures such as standard deviation across runs or keys. These details are load-bearing for assessing whether the cross-device claim is robust.
  2. [Abstract] Abstract: The ~20% outperformance is stated relative to a CNN trained on only 4 devices, while the MLP uses multi-device training. Without a controlled comparison holding the number of training devices fixed (e.g., CNN results with the same multi-device set or MLP results with 4 devices), it is unclear whether the gain is due to the MLP architecture, the PCA step, or simply the larger training distribution.
  3. [Abstract] Abstract: No information is provided on the validation methodology (e.g., whether accuracy is computed per key byte across multiple keys, whether traces are from the same or different plaintexts, or any cross-validation procedure), which is required to evaluate the reliability of the 99.43% and >=10.97% figures.
minor comments (2)
  1. [Abstract] Abstract contains a typographical error: '~20%in terms' is missing a space.
  2. [Abstract] The abstract does not define or cite the specific CNN architecture used as baseline, making the comparison harder to interpret.

Simulated Author's Rebuttal

3 responses · 0 unresolved

Thank you for your review and the valuable feedback on our manuscript. We address each of the major comments below and will update the abstract and add experiments as needed in the revised version.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The 99.43% average accuracy figure is reported without any accompanying information on the total number of traces per device, the train/test split (how many of the 30 devices are used for training versus testing), or statistical measures such as standard deviation across runs or keys. These details are load-bearing for assessing whether the cross-device claim is robust.

    Authors: We agree that these details are important for assessing the claim. The manuscript body describes the dataset from 30 devices along with the multi-device training and testing procedure as well as variability measures. We will revise the abstract to include a concise summary of the number of traces, the train/test split across devices, and statistical measures to make the abstract self-contained. revision: yes

  2. Referee: [Abstract] Abstract: The ~20% outperformance is stated relative to a CNN trained on only 4 devices, while the MLP uses multi-device training. Without a controlled comparison holding the number of training devices fixed (e.g., CNN results with the same multi-device set or MLP results with 4 devices), it is unclear whether the gain is due to the MLP architecture, the PCA step, or simply the larger training distribution.

    Authors: We acknowledge that the reported comparison uses different numbers of training devices and that a controlled experiment would better isolate the sources of improvement. The current results demonstrate the practicality of the MLP+PCA approach when data from many devices is available. We will add a controlled comparison (either CNN on the multi-device set or MLP on 4 devices) in the revised manuscript. revision: yes

  3. Referee: [Abstract] Abstract: No information is provided on the validation methodology (e.g., whether accuracy is computed per key byte across multiple keys, whether traces are from the same or different plaintexts, or any cross-validation procedure), which is required to evaluate the reliability of the 99.43% and >=10.97% figures.

    Authors: We agree that a brief description of the validation approach belongs in the abstract. The reported figures reflect per-key-byte classification accuracy on traces from random plaintexts under a fixed-key-per-device setup with cross-device evaluation. We will revise the abstract to include a short statement on the validation methodology. revision: yes

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The paper reports empirical measurements of classification accuracy on hardware power traces from 30 AVR devices running AES-128. The central result (99.43% average accuracy via PCA pre-processing + multi-device MLP training) follows directly from the described experimental setup and data processing pipeline; no mathematical derivation, prediction step, or uniqueness theorem is claimed that could reduce to fitted parameters or self-citations. The work contains no equations defining quantities in terms of themselves, no renaming of known results as new derivations, and no load-bearing self-citations. This is a standard empirical SCA study whose claims rest on reported test accuracies rather than any closed derivation chain.

Axiom & Free-Parameter Ledger

3 free parameters · 2 axioms · 0 invented entities

The performance claims depend on the effectiveness of standard ML preprocessing and training procedures applied to power trace datasets; no new physical entities or unproven mathematical axioms beyond domain standards.

free parameters (3)
  • PCA dimensionality
    Number of components selected to preprocess traces
  • MLP hyperparameters
    Architecture details tuned for the 256-class classification task
  • Number of training devices = 4+
    Selected for multi-device training to address variations
axioms (2)
  • domain assumption Power traces from AES-128 on AVR devices contain extractable key information
    Core assumption enabling side-channel analysis
  • domain assumption Inter-device variations are primarily linear and capturable by PCA
    Justifies use of PCA for handling device differences

pith-pipeline@v0.9.0 · 5811 in / 1396 out tokens · 30989 ms · 2026-05-25T02:21:18.214205+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

51 extracted references · 51 canonical work pages · 2 internal anchors

  1. [1]

    Differential power analysis,

    P. Kocher, J. Jaffe, and B. Jun, “Differential power analysis,” in Annual International Cryptology Conference . Springer, 1999, pp. 388–397

    work page 1999

  2. [2]

    The em side—channel (s),

    D. Agrawal, B. Archambeault, J. R. Rao, and P. Rohatgi, “The em side—channel (s),” in International Workshop on Cryptographic Hardware and Embedded Systems . Springer, 2002, pp. 29–45

    work page 2002

  3. [3]

    Electromagnetic analysis (ema): Measures and counter-measures for smart cards,

    J.-J. Quisquater and D. Samyde, “Electromagnetic analysis (ema): Measures and counter-measures for smart cards,” in Smart Card Programming and Security. Springer, 2001, pp. 200–210

    work page 2001

  4. [4]

    Electromagnetic analysis: Concrete results,

    K. Gandolfi, C. Mourtel, and F. Olivier, “Electromagnetic analysis: Concrete results,” in International workshop on cryptographic hardware and embedded systems. Springer, 2001, pp. 251–261

    work page 2001

  5. [5]

    Optical time-domain eavesdropping risks of crt displays,

    M. G. Kuhn, “Optical time-domain eavesdropping risks of crt displays,” in Security and Privacy, 2002. Proceedings. 2002 IEEE Symposium on . IEEE, 2002, pp. 3–18

    work page 2002

  6. [6]

    Information leakage from optical emanations,

    J. Loughry and D. A. Umphress, “Information leakage from optical emanations,” ACM Transactions on Information and System Security (TISSEC) , vol. 5, no. 3, pp. 262–289, 2002

    work page 2002

  7. [7]

    Keyboard acoustic emanations,

    D. Asonov and R. Agrawal, “Keyboard acoustic emanations,” in IEEE Symposium on Security and Privacy, 2004. Proceedings. 2004 . IEEE, 2004, pp. 3–11

    work page 2004

  8. [8]

    Template attacks,

    S. Chari, J. R. Rao, and P. Rohatgi, “Template attacks,” in International Workshop on Cryptographic Hardware and Embedded Systems . Springer, 2002, pp. 13–28

    work page 2002

  9. [9]

    Correlation power analysis with a leakage model,

    E. Brier, C. Clavier, and F. Olivier, “Correlation power analysis with a leakage model,” in International workshop on cryptographic hardware and embedded systems . Springer, 2004, pp. 16–29

    work page 2004

  10. [10]

    Convolutional neural networks with data augmentation against jitter-based countermeasures,

    E. Cagli, C. Dumas, and E. Prouff, “Convolutional neural networks with data augmentation against jitter-based countermeasures,” in CHES, 2017, pp. 45–68

    work page 2017

  11. [11]

    Template attacks versus machine learning revisited and the curse of dimensionality in side-channel analysis: extended version,

    L. Lerman, R. Poussier, O. Markowitch, and F.-X. Standaert, “Template attacks versus machine learning revisited and the curse of dimensionality in side-channel analysis: extended version,” Journal of Cryptographic Engineering , vol. 8, no. 4, pp. 301–313, 2018

    work page 2018

  12. [12]

    Breaking mifare desfire mf3icd40: power analysis and templates in the real world,

    D. Oswald and C. Paar, “Breaking mifare desfire mf3icd40: power analysis and templates in the real world,” in CHES, 2011, pp. 207–222

    work page 2011

  13. [13]

    Efficient, portable template attacks,

    M. O. Choudary and M. G. Kuhn, “Efficient, portable template attacks,” IEEE Transactions on Information F orensics and Security , vol. 13, no. 2, pp. 490–501, 2018

    work page 2018

  14. [14]

    Improving cross-device attacks using zero-mean unit-variance normalization,

    D. P. Montminy, R. O. Baldwin, M. A. Temple, and E. D. Laspe, “Improving cross-device attacks using zero-mean unit-variance normalization,” Journal of Cryptographic Engineering , vol. 3, no. 2, pp. 99–110, 2013

    work page 2013

  15. [15]

    Empirical evaluation of multi-device profiling side-channel attacks,

    N. Hanley, M. O’Neill, M. Tunstall, and W. P. Marnane, “Empirical evaluation of multi-device profiling side-channel attacks,” in Signal Processing Systems (SiPS), 2014 IEEE Workshop on . IEEE, 2014

    work page 2014

  16. [16]

    Efficient template attacks based on probabilistic multi-class support vector machines,

    T. Bartkewitz and K. Lemke-Rust, “Efficient template attacks based on probabilistic multi-class support vector machines,” in International Conference on Smart Card Research and Advanced Applications . Springer, 2012, pp. 263–276

    work page 2012

  17. [17]

    Power analysis attack: an approach based on machine learning,

    L. Lerman, G. Bontempi, and O. Markowitch, “Power analysis attack: an approach based on machine learning,” International Journal of Applied Cryptography, vol. 3, no. 2, pp. 97–115, 2014

    work page 2014

  18. [18]

    Study of deep learning techniques for side-channel analysis and introduction to ascad database

    R. Benadjila, E. Prouff, R. Strullu, E. Cagli, and C. Dumas, “Study of deep learning techniques for side-channel analysis and introduction to ascad database.”

    work page

  19. [19]

    Breaking cryptographic implementations using deep learning techniques,

    H. Maghrebi, T. Portigliatti, and E. Prouff, “Breaking cryptographic implementations using deep learning techniques,” in International Conference on Security, Privacy, and Applied Cryptography Engineering . Springer, 2016, pp. 3–26

    work page 2016

  20. [20]

    Profiling power analysis attack based on mlp in dpa contest v4. 2,

    Z. Martinasek, P. Dzurenda, and L. Malina, “Profiling power analysis attack based on mlp in dpa contest v4. 2,” in Telecommunications and Signal Processing (TSP), 2016 39th International Conference on . IEEE, 2016, pp. 223–226

    work page 2016

  21. [21]

    X-deepsca: Cross-device deep learning side channel attack,

    D. Das, A. Golder, J. Danial, S. Ghosh, A. Raychowdhury, and S. Sen, “X-deepsca: Cross-device deep learning side channel attack,” in Proceedings of the 56th Annual Design Automation Conference 2019 . ACM, 2019, p. 134

    work page 2019

  22. [22]

    Deep learning to evaluate secure rsa implementations,

    M. Carbone, V . Conin, M.-A. Cornelie, F. Dassance, G. Dufresne, C. Dumas, E. Prouff, and A. Venelli, “Deep learning to evaluate secure rsa implementations,” Cryptology ePrint Archive, Report 2019/054, 2019, https://eprint.iacr.org/2019/054

    work page 2019

  23. [23]

    Template attacks in principal subspaces,

    C. Archambeau, E. Peeters, F.-X. Standaert, and J.-J. Quisquater, “Template attacks in principal subspaces,” in International Workshop on Cryptographic Hardware and Embedded Systems . Springer, 2006, pp. 1–14

    work page 2006

  24. [24]

    Chipwhisperer: An open-source platform for hardware embedded security research,

    C. O’Flynn and Z. D. Chen, “Chipwhisperer: An open-source platform for hardware embedded security research,” in International Workshop on Constructive Side-Channel Analysis and Secure Design . Springer, 2014, pp. 243–260

    work page 2014

  25. [25]

    Practical template attacks,

    C. Rechberger and E. Oswald, “Practical template attacks,” in International Workshop on Information Security Applications . Springer, 2004, pp. 440–456

    work page 2004

  26. [26]

    Template attacks on masking—resistance is futile,

    E. Oswald and S. Mangard, “Template attacks on masking—resistance is futile,” in Cryptographers’ Track at the RSA Conference . Springer, 2007, pp. 243–256

    work page 2007

  27. [27]

    A machine learning approach against a masked aes,

    L. Lerman, G. Bontempi, and O. Markowitch, “A machine learning approach against a masked aes,” Journal of Cryptographic Engineering , vol. 5, no. 2, pp. 123–139, 2015

    work page 2015

  28. [28]

    Intelligent machine homicide,

    A. Heuser and M. Zohner, “Intelligent machine homicide,” in International Workshop on Constructive Side-Channel Analysis and Secure Design . Springer, 2012, pp. 249–264

    work page 2012

  29. [29]

    A time series approach for profiling attack,

    L. Lerman, G. Bontempi, S. B. Taieb, and O. Markowitch, “A time series approach for profiling attack,” in International Conference on Security, Privacy, and Applied Cryptography Engineering . Springer, 2013, pp. 75–94

    work page 2013

  30. [30]

    Neural network based attack on a masked implementation of aes,

    R. Gilmore, N. Hanley, and M. O’Neill, “Neural network based attack on a masked implementation of aes,” in IEEE HOST , 2015, pp. 106–111

    work page 2015

  31. [31]

    Optimization of power analysis using neural network,

    Z. Martinasek, J. Hajny, and L. Malina, “Optimization of power analysis using neural network,” in International Conference on Smart Card Research and Advanced Applications . Springer, 2013, pp. 94–107

    work page 2013

  32. [32]

    A formal study of power variability issues and side-channel attacks for nanoscale devices,

    M. Renauld, F.-X. Standaert, N. Veyrat-Charvillon, D. Kamel, and D. Flandre, “A formal study of power variability issues and side-channel attacks for nanoscale devices,” in Annual International Conference on the Theory and Applications of Cryptographic Techniques . Springer, 2011, pp. 109–128

    work page 2011

  33. [33]

    Principal component analysis,

    I. Jolliffe, “Principal component analysis,” in International encyclopedia of statistical science . Springer, 2011, pp. 1094–1096

    work page 2011

  34. [34]

    Dynamic time warping,

    M. M ¨uller, “Dynamic time warping,” Information retrieval for music and motion , pp. 69–84, 2007

    work page 2007

  35. [35]

    Dpa contest , 2008–2009

    T. P. S. research group et al. , “Dpa contest , 2008–2009.”

    work page 2008

  36. [36]

    Dpa contest , 2013–2014

    T. P. S. R. Group et al. , “Dpa contest , 2013–2014.”

    work page 2013

  37. [37]

    Make some noise: Unleashing the power of convolutional neural networks for profiled side-channel analysis,

    J. Kim, S. Picek, A. Heuser, S. Bhasin, and A. Hanjalic, “Make some noise: Unleashing the power of convolutional neural networks for profiled side-channel analysis,” Cryptology ePrint Archive, Report 2018/1023, 2018, https://eprint.iacr.org/2018/1023

    work page 2018

  38. [38]

    Convolutional neural network based side-channel attacks in time-frequency representations,

    G. Yang, H. Li, J. Ming, and Y . Zhou, “Convolutional neural network based side-channel attacks in time-frequency representations,” in International Conference on Smart Card Research and Advanced Applications . Springer, 2018, pp. 1–17

    work page 2018

  39. [39]

    Improving side-channel analysis through semi-supervised learning,

    S. Picek, A. Heuser, A. Jovic, K. Knezevic, and T. Richmond, “Improving side-channel analysis through semi-supervised learning,” in International Conference on Smart Card Research and Advanced Applications . Springer, 2018, pp. 35–50

    work page 2018

  40. [40]

    Loss functions for discriminative training of energy-based models

    Y . LeCun and F. J. Huang, “Loss functions for discriminative training of energy-based models.” in AIStats, vol. 6, 2005, p. 34

    work page 2005

  41. [41]

    Adam: A Method for Stochastic Optimization

    D. P. Kingma and J. Ba, “Adam: A method for stochastic optimization,” arXiv preprint arXiv:1412.6980 , 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  42. [42]

    Getting more from pca: first results of using principal component analysis for extensive power analysis,

    L. Batina, J. Hogenboom, and J. G. van Woudenberg, “Getting more from pca: first results of using principal component analysis for extensive power analysis,” in RSA, 2012, pp. 383–397

    work page 2012

  43. [43]

    Efficient stochastic methods: profiled attacks beyond 8 bits,

    M. O. Choudary and M. G. Kuhn, “Efficient stochastic methods: profiled attacks beyond 8 bits,” in International Conference on Smart Card Research and Advanced Applications . Springer, 2014, pp. 85–103

    work page 2014

  44. [44]

    Enhancing dimensionality reduction methods for side-channel attacks,

    E. Cagli, C. Dumas, and E. Prouff, “Enhancing dimensionality reduction methods for side-channel attacks,” in International Conference on Smart Card Research and Advanced Applications . Springer, 2015, pp. 15–33

    work page 2015

  45. [45]

    Improving differential power analysis by elastic alignment,

    J. G. van Woudenberg, M. F. Witteman, and B. Bakker, “Improving differential power analysis by elastic alignment,” in Cryptographers’ Track at the RSA Conference. Springer, 2011, pp. 104–119

    work page 2011

  46. [46]

    Evaluation of dynamic voltage and frequency scaling as a differential power analysis countermeasure,

    K. Baddam and M. Zwolinski, “Evaluation of dynamic voltage and frequency scaling as a differential power analysis countermeasure,” in IEEE VLSI Design, 2007, pp. 854–862

    work page 2007

  47. [47]

    Rijid: random code injection to mask power analysis based side channel attacks,

    J. A. Ambrose, R. G. Ragel, and S. Parameswaran, “Rijid: random code injection to mask power analysis based side channel attacks,” in DAC, 2007, pp. 489–492

    work page 2007

  48. [48]

    Dynamic programming algorithm optimization for spoken word recognition,

    H. Sakoe and S. Chiba, “Dynamic programming algorithm optimization for spoken word recognition,” IEEE transactions on acoustics, speech, and signal processing, vol. 26, no. 1, pp. 43–49, 1978

    work page 1978

  49. [49]

    Keras: The python deep learning library,

    F. Chollet et al. , “Keras: The python deep learning library,” Astrophysics Source Code Library , 2018

    work page 2018

  50. [50]

    Tensorflow: a system for large-scale machine learning

    M. Abadi, P. Barham, J. Chen, Z. Chen, A. Davis, J. Dean, M. Devin, S. Ghemawat, G. Irving, M. Isard et al. , “Tensorflow: a system for large-scale machine learning.” in OSDI, vol. 16, 2016, pp. 265–283

    work page 2016

  51. [51]

    Very Deep Convolutional Networks for Large-Scale Image Recognition

    K. Simonyan and A. Zisserman, “Very deep convolutional networks for large-scale image recognition,” arXiv preprint arXiv:1409.1556 , 2014

    work page internal anchor Pith review Pith/arXiv arXiv 2014