Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 112 of the 112 references it cites were found in a registry or an index; 18 of the 30 sources its claims cite were checked for support.

EEGNet: a compact convolutional neural network for EEG-based brain-computer interfaces

Open access
Lawhern, Vernon J, Solon, Amelia J, Waytowich, Nicholas R, Gordon, Stephen M, Hung, Chou P, Lance, Brent J · 2018 · Journal of Neural Engineering

Abstract

OBJECTIVE: Brain-computer interfaces (BCI) enable direct communication with a computer, using neural activity as the control signal. This neural signal is generally chosen from a variety of well-studied electroencephalogram (EEG) signals. For a given BCI paradigm, feature extractors and classifiers are tailored to the distinct characteristics of its expected EEG control signal, limiting its application to that specific signal. Convolutional neural networks (CNNs), which have been used in computer vision and speech recognition to perform automatic feature extraction and classification, have successfully been applied to EEG-based BCIs; however, they have mainly been applied to single BCI paradigms and thus it remains unclear how these architectures generalize to other paradigms. Here, we ask if we can design a single CNN architecture to accurately classify EEG signals from different BCI paradigms, while simultaneously being as compact as possible. APPROACH: In this work we introduce EEGNet, a compact convolutional neural network for EEG-based BCIs. We introduce the use of depthwise and separable convolutions to construct an EEG-specific model which encapsulates well-known EEG feature extraction concepts for BCI. We compare EEGNet, both for within-subject and cross-subject classification, to current state-of-the-art approaches across four BCI paradigms: P300 visual-evoked potentials, error-related negativity responses (ERN), movement-related cortical potentials (MRCP), and sensory motor rhythms (SMR). MAIN RESULTS: We show that EEGNet generalizes across paradigms better than, and achieves comparably high performance to, the reference algorithms when only limited training data is available across all tested paradigms. In addition, we demonstrate three different approaches to visualize the contents of a trained EEGNet model to enable interpretation of the learned features. SIGNIFICANCE: Our results suggest that EEGNet is robust enough to learn a wide variety of interpretable features over a range of BCI tasks. Our models can be found at: https://github.com/vlawhern/arl-eegmodels.

Overview

CoChat Summary

TL;DR

EEGNet is a compact CNN using depthwise and separable convolutions that generalizes across P300, ERN, MRCP, and SMR BCI paradigms while using far fewer parameters than reference CNNs. It achieves comparable or better performance than state-of-the-art approaches in many within- and cross-subject comparisons and produces neurophysiologically interpretable features.

Key findings

  • EEGNet models were up to two orders of magnitude smaller than DeepConvNet and ShallowConvNet: e.g., P300 EEGNet-4,2 had 1,066 parameters vs. 174,127 for DeepConvNet and 104,002 for ShallowConvNet; SMR EEGNet-4,2 had 796 vs. 152,219 and 40,644, respectively.
  • Within-subject, EEGNet-4,2 and EEGNet-8,2 did not differ significantly across paradigms (p > 0.05). Both EEGNet models significantly outperformed all other models on MRCP (p < 0.05); EEGNet-8,2 outperformed DeepConvNet, ShallowConvNet, and xDAWN+RG on ERN (p < 0.05); all CNN models outperformed xDAWN+RG on P300 (p < 0.05).
  • Cross-subject, all CNN models outperformed xDAWN+RG on P300 (p < 0.05); EEGNet-8,2 and DeepConvNet outperformed ShallowConvNet on MRCP (p < 0.05); xDAWN+RG outperformed all CNN models on ERN (p < 0.05); no significant CNN differences were found for SMR (p > 0.05).
  • Feature analyses showed interpretable structure: P300 ablation gave full-model AUC 0.9054, with removal of filters (1,3,4) reducing AUC to 0.7108; SMR spatial filters learned by EEGNet-8,2 correlated with FBCSP filters (e.g., ρ = 0.93, ρ = 0.83, and ρ = −0.93).

Methods

  • EEGNet uses temporal convolution, depthwise convolution for frequency-specific spatial filters, and separable convolution, with batch normalization, ELU, dropout, average pooling, and softmax classification; trained with Adam, 500 epochs, validation stopping.
  • Evaluated on four datasets: P300 (15 subjects, ~2000 trials/subject, 1–40 Hz), ERN (26 subjects, 340 trials/subject, 1–40 Hz), MRCP (13 subjects, ~1100 trials/subject, 0.1–40 Hz), and SMR (9 subjects, 288 trials/subject, 4–40 Hz).
  • Within-subject used four-fold blockwise cross-validation; cross-subject used repeated random folds (30 for P300/MRCP/ERN, 90 for SMR). Comparisons included DeepConvNet, ShallowConvNet, xDAWN+RG for ERP datasets, and FBCSP for SMR, with repeated-measures or one-way ANOVA.

Limitations

  • The paper does not state a formal limitations section. Evident from the text: validation is limited to four BCI paradigms/datasets, and cross-subject ERN results show xDAWN+RG significantly outperforming all CNN models.
  • The comparison to xDAWN+RG used a single model rather than the original ensemble and omitted dataset-specific “meta features,” as stated by the authors.
  • The authors note that ShallowConvNet’s severe underperformance on within-subject MRCP was unexpected and that determining its underlying source will be explored in future research.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1088/1741-2552/aace8c

Cites

112 of 112 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

8 Valid · 1 Not valid · 2 Not checked · 61 Own results and interpretations

35 more claims

Create a free account to see all 72 claims and to verify the 1 reference still open.

See all claims — free account
  • It may state two separate claims as one. 20
  • Probably not a substantial claim. 15

Create a free account to see these claims.

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

17 more claims about this paper

Create a free account to see every claim other papers make about this one.

See all claims — free account

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 78 references

Research works printed but not linked: 21

  1. Zander T O, 2011, J. Neural Eng., 8
  2. Krizhevsky A, 2012, Advances in Neural Information Processing Systems, 1097
  3. Simonyan K, 2014, CoRR
  4. Szegedy C, 2014, CoRR
  5. He K, 2015, CoRR
  6. Huang G, 2016, CoRR
  7. Thodoroff P, 2016, CoRR
  8. Stober S, 2014, Advances in Neural Information Processing Systems 27, 1449
  9. Stober S, 2015, CoRR
  10. Bashivan P, 2015, CoRR
  11. Chollet F, 2016, CoRR
  12. Abadi M, 2016, Proc. 12th USENIX Conf. on Operating Systems Design and Implementation, 265
  13. Dyrholm M, 2007, J. Mach. Learn. Res., 8, 1097
  14. Clevert D, 2015, CoRR
  15. Srivastava N, 2014, J. Mach. Learn. Res., 15, 1929
  16. Congedo M, 2013, CoRR
  17. Baehrens D, 2010, J. Mach. Learn. Res., 11, 1803
  18. Nguyen A M, 2014, CoRR
  19. Shrikumar A, 2017, CoRR
  20. Ancona M, 2018, Int. Conf. on Learning Representations
  21. 10.1175/1520-0477(1998)079<0061:APGTWA>2.0.CO;2 · DOI · Paper page

Cited by

See all 5,115

Metadata

Authors
Lawhern, V. J., Solon, A. J., Waytowich, N. R., Gordon, S. M., Hung, C. P., & Lance, B. J.
Affiliations
DEVCOM Army Research Laboratory (US)
Corresponding author
Lawhern, Vernon J
Venue
Journal of Neural Engineering
Published
2018-07-27
Volume
15
Issue
5
Pages
056013
ISSN
  • 1741-2552 (ISSN-L)
  • 1741-2560
OpenAlex
W2559463885
Cited by
5,115
References
78
Citations by year
2026: 1132 · 2025: 1233 · 2024: 935 · 2023: 649 · 2022: 468 · 2021: 334 · 2020: 212 · 2019: 105 · 2018: 37
FWCI
102.30
Type
journal article
Language
en
Open access
Licence
Locations
  • Journal of Neural Engineering (publishedVersion)
  • PubMed (publishedVersion)
  • arXiv (Cornell University) (submittedVersion)
  • arXiv (Cornell University)
Funders
U.S. Army Research Laboratory (CAST 076910227001, ARL-74A-HRCYB, ARL-H70-HR52, W911NF-10-2-0022)
Topics
  • Life Sciences › Neuroscience › Cognitive Neuroscience › EEG and Brain-Computer Interfaces
  • Physical Sciences › Computer Science › Artificial Intelligence › Emotion and Mood Recognition
  • Physical Sciences › Computer Science › Human-Computer Interaction › Gaze Tracking and Assistive Technology
Keywords
EEGNet, EEG-based brain-computer interface, compact convolutional neural networks, depthwise separable convolution, error-related negativity, movement-related cortical potentials, sensory motor rhythm, EEG classification, model interpretability, feature visualization, limited training data
MeSH terms
Adolescent, Adult, Algorithms, Electroencephalography, Evoked Potentials, Visual, Female, Humans, Male, Middle Aged, Movement, Neural Networks, Computer, Event-Related Potentials, P300, Young Adult, Brain-Computer Interfaces

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Cited sources are checked with a language model’s help (qwen3.8-omni-flash and gpt-6-sol).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: EEGNet: a compact convolutional neural network for EEG-based brain-computer interfaces. https://app.cochat.ai/papers/10.1088/1741-2552/aace8c
The paper
Lawhern, V. J., Solon, A. J., Waytowich, N. R., Gordon, S. M., Hung, C. P., & Lance, B. J. (2018). EEGNet: a compact convolutional neural network for EEG-based brain-computer interfaces. Journal of Neural Engineering. https://doi.org/10.1088/1741-2552/aace8c

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.