pith. sign in

arxiv: 1907.05674 · v1 · pith:6DBFHBWTnew · submitted 2019-07-12 · 📡 eess.SP · cs.HC· cs.LG

Deep Learning with ConvNET Predicts Imagery Tasks Through EEG

Apdullah Yay{\i}k , Yakup Kutlu , G\"okhan Altan This is my paper

Pith reviewed 2026-05-24 22:23 UTC · model grok-4.3

classification 📡 eess.SP cs.HCcs.LG
keywords EEGconvolutional neural networksmotor imagerydeep learningsubject-independent classificationbrain signal analysis
0
0 comments X

The pith

ConvNets on raw EEG beat spectral features for motor imagery prediction

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

The paper aims to establish that convolutional neural networks can classify imagined left and right hand movements directly from raw EEG signals on a subject-independent basis. It shows that adding adaptive moment estimation, batch normalization, and dropout improves results over standard fully connected networks that rely on spectral features. A sympathetic reader would care because this points to deep learning bypassing manual feature design for brain-signal tasks, which could simplify analysis in applications like brain-computer interfaces.

Core claim

ConvNets of different structures, constructed for predicting imagined left and right movements on a subject-independent basis through raw EEG data, achieve improved prediction when using adaptive moments, batch normalization, and dropout, outperforming conventional fully-connected neural networks with spectral features.

What carries the argument

Convolutional neural networks applied directly to raw EEG signals, trained with adaptive moment estimation, batch normalization, and dropout.

If this is right

  • ConvNets learn task-relevant patterns from raw EEG without needing prior spectral feature extraction.
  • Subject-independent prediction of imagined movements becomes feasible with these networks.
  • Modern optimization and regularization methods increase ConvNet effectiveness on EEG data.
  • The approach exceeds performance of traditional neural networks that depend on hand-engineered spectral features.

Where Pith is reading between the lines

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

  • If the gains hold across datasets, EEG decoding systems could require less expert input for feature selection.
  • The same raw-data ConvNet strategy may extend to other EEG classification problems such as emotion detection.
  • Further tests on varied recording conditions would show how robust the subject-independent results remain.

Load-bearing premise

The EEG recordings used contain sufficient subject-independent signal structure that the reported performance gain can be attributed to the ConvNet architecture and training choices rather than dataset idiosyncrasies or unstated preprocessing steps.

What would settle it

Applying the same ConvNet architectures and training methods to an independent EEG motor imagery dataset and finding no performance advantage over spectral-feature baselines would falsify the claim.

read the original abstract

Deep learning with convolutional neural networks (ConvNets) have dramatically improved learning capabilities of computer vision applications just through considering raw data without any prior feature extraction. Nowadays, there is rising curiosity in interpreting and analyzing electroencephalography (EEG) dynamics with ConvNets. Our study focused on ConvNets of different structures, constructed for predicting imagined left and right movements on a subject-independent basis through raw EEG data. Results showed that recently advanced methods in machine learning field, i.e. adaptive moments and batch normalization together with dropout strategy, improved ConvNets predicting ability, outperforming that of conventional fully-connected neural networks with widely-used spectral features.

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

2 major / 1 minor

Summary. The manuscript describes the application of convolutional neural networks (ConvNets) to raw EEG signals for classifying imagined left and right hand movements on a subject-independent basis. It reports that combining adaptive moment estimation (Adam), batch normalization, and dropout improves the predictive performance of ConvNets compared to fully-connected neural networks using spectral features.

Significance. If the reported performance improvements are validated with appropriate subject-independent protocols and quantitative metrics, the work could contribute to the development of brain-computer interfaces by demonstrating the effectiveness of modern deep learning techniques on unprocessed EEG data without relying on hand-engineered features.

major comments (2)
  1. [Abstract] Abstract: No quantitative accuracies, dataset sizes, number of subjects, cross-validation details, error bars, or baseline comparisons are provided. This makes it impossible to evaluate the central claim that the ConvNet with Adam/BN/dropout outperforms conventional methods.
  2. [Abstract] Abstract: The claim of subject-independent prediction requires an explicit description of the evaluation protocol (e.g., leave-one-subject-out cross-validation) to ensure no data leakage from subject-specific correlations; none is supplied, which is load-bearing for attributing gains to the architecture and training choices rather than dataset artifacts.
minor comments (1)
  1. [Abstract] Grammatical issues: 'Deep learning with convolutional neural networks (ConvNets) have dramatically' should be 'has'; 'predicting ability' could be 'predictive performance'.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments on our manuscript. We agree that the abstract requires additional quantitative details and an explicit description of the subject-independent evaluation protocol to strengthen the presentation of our results. We will revise the abstract in the next version of the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: No quantitative accuracies, dataset sizes, number of subjects, cross-validation details, error bars, or baseline comparisons are provided. This makes it impossible to evaluate the central claim that the ConvNet with Adam/BN/dropout outperforms conventional methods.

    Authors: We acknowledge that the current abstract does not include specific numerical results or experimental details. In the revised version, we will incorporate key quantitative metrics (accuracies, dataset size, number of subjects, cross-validation scheme, error bars, and baseline comparisons) directly into the abstract to allow readers to evaluate the performance claims. revision: yes

  2. Referee: [Abstract] Abstract: The claim of subject-independent prediction requires an explicit description of the evaluation protocol (e.g., leave-one-subject-out cross-validation) to ensure no data leakage from subject-specific correlations; none is supplied, which is load-bearing for attributing gains to the architecture and training choices rather than dataset artifacts.

    Authors: We agree that the abstract should explicitly state the subject-independent evaluation protocol. We will add a concise description of the cross-validation procedure (including how subject independence is enforced) to the abstract to clarify that gains are not due to data leakage. revision: yes

Circularity Check

0 steps flagged

No circularity: purely empirical ML evaluation with no derivation chain

full rationale

The paper reports experimental results from training and evaluating ConvNets on EEG motor imagery data. No mathematical derivations, first-principles claims, or predictions are made that could reduce to fitted parameters or self-citations by construction. Performance comparisons are presented as direct empirical outcomes rather than derived quantities. The evaluation protocol details are unspecified in the provided text, but this is a methodological limitation, not a circularity issue.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities are described in the provided text.

pith-pipeline@v0.9.0 · 5644 in / 967 out tokens · 22985 ms · 2026-05-24T22:23:24.548666+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

24 extracted references · 24 canonical work pages · 2 internal anchors

  1. [1]

    A cell-phone- based brain–computer interface for communication in daily life

    Yu-Te Wang, Yijun Wang, and Tzyy-Ping Jung. A cell-phone- based brain–computer interface for communication in daily life. Journal of neural engineering, 8(2):025018, 2011. [1]

    work page 2011

  2. [2]

    Brain-controlledwheelchairs: a robotic architecture

    TomCarlsonandJosedelRMillan. Brain-controlledwheelchairs: a robotic architecture. IEEE Robotics & Automation Magazine , 20(1):65–73, 2013. [1]

    work page 2013

  3. [3]

    A brain computer interface-basedexplorer

    Lijuan Bai, Tianyou Yu, and Yuanqing Li. A brain computer interface-basedexplorer. Journal of neuroscience methods,244:2– 7, 2015. [1]

    work page 2015

  4. [4]

    Patient specific seizure pre- dictionsystemusinghilbertspectrumandbayesiannetworksclas- sifiers

    Nilufer Ozdemir and Esen Yildirim. Patient specific seizure pre- dictionsystemusinghilbertspectrumandbayesiannetworksclas- sifiers. Computational and mathematical methods in medicine , 2014, 2014. [1]

    work page 2014

  5. [5]

    Deep learn- ing

    IanGoodfellow,YoshuaBengio,andAaronCourville. Deep learn- ing. MIT press, 2016. [1]

    work page 2016

  6. [6]

    Deep convolutional neural networks for sentiment analysis of short texts

    Cicero Dos Santos and Maira Gatti. Deep convolutional neural networks for sentiment analysis of short texts. InProceedings of COLING 2014, the 25th International Conference on Computa- tional Linguistics: Technical Papers, pages 69–78, 2014. [1]

    work page 2014

  7. [7]

    Deep neural networks for acoustic modeling in speech recognition.IEEE Signal processing magazine, 29, 2012

    Geoffrey Hinton, Li Deng, Dong Yu, George Dahl, Abdel-rahman Mohamed, Navdeep Jaitly, Andrew Senior, Vincent Vanhoucke, Patrick Nguyen, Brian Kingsbury, et al. Deep neural networks for acoustic modeling in speech recognition.IEEE Signal processing magazine, 29, 2012. [1]

    work page 2012

  8. [8]

    Classification of executed and imagined motor movement eeg signals.Ann Arbor: University of Michigan, pages 1–10, 2009

    Jason Sleight, Preeti Pillai, and Shiwali Mohan. Classification of executed and imagined motor movement eeg signals.Ann Arbor: University of Michigan, pages 1–10, 2009. [1]

    work page 2009

  9. [9]

    Time- frequency analysis of executed and imagined motor movement eeg signals for neuro-based home appliance system

    MHF Zakaria, Wahidah Mansor, and Khuan Y Lee. Time- frequency analysis of executed and imagined motor movement eeg signals for neuro-based home appliance system. InTENCON 2017-2017 IEEE Region 10 Conference, pages 1657–1660. IEEE,

    work page 2017

  10. [10]

    Eeg-based classification of imagined fists movements using ma- chinelearningandwavelettransformanalysis

    Mohammad H Alomari, Ali M Baniyounes, and Emad A Awada. Eeg-based classification of imagined fists movements using ma- chinelearningandwavelettransformanalysis. Int. J. Adv. Electron. Electr. Eng, pages 83–87, 2014. [1, 4]

    work page 2014

  11. [11]

    An eeg pre-processing technique for the fast recognition of mo- tor imagery movements

    Kalogiannis Gregory, Kapsimanis George, and Hassapis George. An eeg pre-processing technique for the fast recognition of mo- tor imagery movements. In2016 IEEE Biomedical Circuits and Systems Conference (BioCAS), pages 90–94. IEEE, 2016. [1]

    work page 2016

  12. [12]

    Multiclass common spatial pattern for eeg based brain computer interface withadaptivelearningclassifier

    Hardik Meisheri, Nagraj Ramrao, and Suman Mitra. Multiclass common spatial pattern for eeg based brain computer interface withadaptivelearningclassifier. arXiv preprint arXiv:1802.09046,

    work page arXiv

  13. [13]

    Eeg-based intention recognition from spatio- temporal representations via cascade and parallel convolutional recurrentneuralnetworks

    Dalin Zhang, Lina Yao, Xiang Zhang, Sen Wang, Weitong Chen, and Robert Boots. Eeg-based intention recognition from spatio- temporal representations via cascade and parallel convolutional recurrentneuralnetworks. arXiv preprint arXiv:1708.06578,2017. [2]

    work page arXiv 2017

  14. [14]

    Dropout: a simple way to prevent neural networks from overfitting.The Journal of Machine Learning Research, 15(1):1929–1958, 2014

    Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever, and Ruslan Salakhutdinov. Dropout: a simple way to prevent neural networks from overfitting.The Journal of Machine Learning Research, 15(1):1929–1958, 2014. [2]

    work page 1929

  15. [15]

    Batch Normalization: Accelerating Deep Network Training by Reducing Internal Covariate Shift

    Sergey Ioffe and Christian Szegedy. Batch normalization: Accel- erating deep network training by reducing internal covariate shift. arXiv preprint arXiv:1502.03167, 2015. [2, 3]

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  16. [16]

    Neural networks for machine learning lecture 6a overview of mini-batch gradient descent

    Geoffrey Hinton, Nitish Srivastava, and Kevin Swersky. Neural networks for machine learning lecture 6a overview of mini-batch gradient descent. Cited on, 14, 2012. [2]

    work page 2012

  17. [17]

    Adam: A Method for Stochastic Optimization

    DiederikPKingmaandJimmyBa. Adam:Amethodforstochastic optimization. arXiv preprint arXiv:1412.6980, 2014. [2]

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  18. [18]

    Bci2000: a general-purpose brain-computer interface (bci) system

    Gerwin Schalk, Dennis J McFarland, Thilo Hinterberger, Niels Birbaumer, and Jonathan R Wolpaw. Bci2000: a general-purpose brain-computer interface (bci) system. IEEE Transactions on biomedical engineering, 51(6):1034–1043, 2004. [2]

    work page 2004

  19. [19]

    Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals.Circulation, 101(23):e215–e220, 2000

    Ary L Goldberger, Luis AN Amaral, Leon Glass, Jeffrey M Hausdorff, Plamen Ch Ivanov, Roger G Mark, Joseph E Mietus, George B Moody, Chung-Kang Peng, and H Eugene Stanley. Physiobank, physiotoolkit, and physionet: components of a new research resource for complex physiologic signals.Circulation, 101(23):e215–e220, 2000. [2]

    work page 2000

  20. [20]

    The use of fast fourier transform for the estimation of power spectra: a method based on time averaging over short, modified periodograms

    Peter Welch. The use of fast fourier transform for the estimation of power spectra: a method based on time averaging over short, modified periodograms. IEEE Transactions on audio and elec- troacoustics, 15(2):70–73, 1967. [2] Deep Learning with ConvNET Predicts Imagery Tasks Through EEG 5 (a) (b) (c) (d) Fig. 2 Confusion matrices for (a) Mutilayer perceptr...

    work page 1967

  21. [21]

    Analyzing neural time series data: theory and practice

    Mike X Cohen. Analyzing neural time series data: theory and practice. MIT press, 2014. [2]

    work page 2014

  22. [22]

    Neocognitron: A self- organizingneuralnetworkmodelforamechanismofvisualpattern recognition

    Kunihiko Fukushima and Sei Miyake. Neocognitron: A self- organizingneuralnetworkmodelforamechanismofvisualpattern recognition. InCompetition and cooperation in neural nets,pages 267–285. Springer, 1982. [3]

    work page 1982

  23. [23]

    Gradient-based learning applied to document recognition

    Yann LeCun, Léon Bottou, Yoshua Bengio, Patrick Haffner, et al. Gradient-based learning applied to document recognition. Pro- ceedings of the IEEE, 86(11):2278–2324, 1998. [3]

    work page 1998

  24. [24]

    Deep learning with convolutional neural networks for eeg decoding and visualization.Human Brain Map- ping, aug 2017

    Robin Tibor Schirrmeister, Jost Tobias Springenberg, Lukas Dominique Josef Fiederer, Martin Glasstetter, Katharina Eggensperger, Michael Tangermann, Frank Hutter, Wolfram Bur- gard, and Tonio Ball. Deep learning with convolutional neural networks for eeg decoding and visualization.Human Brain Map- ping, aug 2017. [3]

    work page 2017