CoChat

Low dimensional subject representation based transfer learning in eeg decoding

Open access
Jeng, Po-Yuan, Wei, Chun-Shu, Jung, Tzyy-Ping, Wang, Li-Chun · 2021 · IEEE Journal of Biomedical and Health Informatics

The DOI matches its Crossref record; 54 of the 54 references it cites were found in a registry or an index.

Abstract

Recently, the advances in passive brain-computer interfaces (BCIs) based on electroencephalogram (EEG) have shed light on real-world neuromonitoring technologies. However, human variability in the EEG activities hinders the development of practical applications of EEG-based BCI. To tackle this problem, many transfer-learning techniques perform supervised calibration. This kind of calibration approach requires task-relevant data, which is impractical in real-life scenarios such as drowsiness during driving. This study presents a transfer-learning framework for EEG decoding based on the low-dimensional representations of subjects learned from the pre-trial EEG. Tensor decomposition was applied to the pre-trial EEG of subjects to extract the underlying characteristics in subject, spatial, and spectral domains. Then, the proposed framework assessed the characteristics to obtain the low-dimensional subject representations such that the subjects with similar brain dynamics can be identified. This method can leverage the existing data from other users, and a small number of data from a rapid, non-task, unsupervised calibration from a new user to build an accurate BCI. Our results demonstrated that, in terms of prediction accuracy, the proposed low-dimensional subject representation-based transfer learning (LDSR-TL) framework outperformed the random selection, and the Riemannian manifold approach in cognitive-state tracking, while requiring fewer training data. The results can greatly improve the practicability, and usability of EEG-based BCI in the real world.

Overview

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

Integrity signals

Registry

Crossref · 10.1109/jbhi.2020.3025865

Cites

54 of 54 references verified0 citation errors · 0 not checkable

Statistics

Figures

Fig. 1 Related works categorization in a binary tree.
Fig. 1 Related works categorization in a binary tree.CC-BY-NC-ND · From the PDF
Image 2
CC-BY-NC-ND · From the PDF
Fig. 3 Tensor decomposition decompose a session into weights of frequency-channel interactions that are the components of composing other sessions.
Fig. 3 Tensor decomposition decompose a session into weights of frequency-channel interactions that are the components of composing other sessions.CC-BY-NC-ND · From the PDF
Fig. 4 Correlation of latent representations of sessions/subjects visualized on a heat map. The brighter color represents higher correlations between two sessions. The sessions from the same subject are more similar to each other.
Fig. 4 Correlation of latent representations of sessions/subjects visualized on a heat map. The brighter color represents higher correlations between two sessions. The sessions from the same subject are more similar to each other.CC-BY-NC-ND · From the PDF
Fig. 5 The flow diagram of the proposed subject-transfer framework.
Fig. 5 The flow diagram of the proposed subject-transfer framework.CC-BY-NC-ND · From the PDF
Fig. 6 The pre-trial EEG are stacked into a tensor and then decomposed to obtain low-dimensional subject representations. First, the pre-trial EEG PSD data are tensorized. The three modes of the tensor are the subjects, channels and frequency bins. An element of X ijk is the Log-power measurement of the k th frequency bin at j th channel of the i th subject. Then, tensor decomposition is applied to the tensor for learning the low-dimensional representations of subjects.
Fig. 6 The pre-trial EEG are stacked into a tensor and then decomposed to obtain low-dimensional subject representations. First, the pre-trial EEG PSD data are tensorized. The three modes of the tensor are the subjects, channels and frequency bins. An element of X ijk is the Log-power measurement of the k th frequency bin at j th channel of the i th subject. Then, tensor decomposition is applied to the tensor for learning the low-dimensional representations of subjects.CC-BY-NC-ND · From the PDF
Fig. 7 Overview of the driving experiment. Participants must drive the car and keep it in-lane by steering the steering wheel. Multiple drift events occur randomly, and the subject must steer the car back to the cruising position. The time elapsed between drift onset and subject steering back is called the reaction time. The EEG in the three-seconds window will be recorded with the reaction time as a trial.
Fig. 7 Overview of the driving experiment. Participants must drive the car and keep it in-lane by steering the steering wheel. Multiple drift events occur randomly, and the subject must steer the car back to the cruising position. The time elapsed between drift onset and subject steering back is called the reaction time. The EEG in the three-seconds window will be recorded with the reaction time as a trial.CC-BY-NC-ND · From the PDF
Fig. 8 The spectrogram of sorted PSD trials, showing that there exists a relation between PSD and reaction time in some but not all of the trials.
Fig. 8 The spectrogram of sorted PSD trials, showing that there exists a relation between PSD and reaction time in some but not all of the trials.CC-BY-NC-ND · From the PDF
Fig. 9 The overview of our LOSO cross-validation. Each subject was designated as the target once, and other subjects were added to the source pool. If multiple sessions were recorded from the same subject, each session was set as the target once and other sessions from the same subject were excluded from the source pool.
Fig. 9 The overview of our LOSO cross-validation. Each subject was designated as the target once, and other subjects were added to the source pool. If multiple sessions were recorded from the same subject, each session was set as the target once and other sessions from the same subject were excluded from the source pool.CC-BY-NC-ND · From the PDF
Fig. 10 Performance comparison of the proposed method and two baselines under using Bayesian Regression as regressor. These curves show the overall performance in the DI prediction against the number of transferred sessions that are selected by the proposed method and others. In general, the performance of all three approaches grows with more sessions being selected and transferred. Significant differences ( p < 0 . 05 ) in the overall performance assessed by paired t-test were marked by '*' (LDSR-TL VS random) and ' # ' (LDSR-TL vs [9]). Our method showed its superiority when the training data were limited (8 selected sessions).
Fig. 10 Performance comparison of the proposed method and two baselines under using Bayesian Regression as regressor. These curves show the overall performance in the DI prediction against the number of transferred sessions that are selected by the proposed method and others. In general, the performance of all three approaches grows with more sessions being selected and transferred. Significant differences ( p < 0 . 05 ) in the overall performance assessed by paired t-test were marked by '*' (LDSR-TL VS random) and ' # ' (LDSR-TL vs [9]). Our method showed its superiority when the training data were limited (8 selected sessions).CC-BY-NC-ND · From the PDF
Fig. 11 A comparison of the overall drowsiness-prediction performance across different regressors. The number of selected sessions was fixed at eight. Significant differences (* p < 0 . 05 ) in the overall performance were assessed by paired t-test.
Fig. 11 A comparison of the overall drowsiness-prediction performance across different regressors. The number of selected sessions was fixed at eight. Significant differences (* p < 0 . 05 ) in the overall performance were assessed by paired t-test.CC-BY-NC-ND · From the PDF
Image 12
CC-BY-NC-ND · From the PDF
Fig. 13 Prediction results of proposed and baseline methods for the target session, S5-3. The red line is the predicted DI by the proposed method. The blue line is predicted by [9]. The gray line is predicted by the random selection method. The black line is the ground truth, the DI derived from the subject's behavioral responses. The correlations between the actual and predicted DIs are also listed. The proposed method selected the subjects who have similar EEG characteristics for transfer learning, so that the trained model was able to recognize drowsy episodes of the new subject.
Fig. 13 Prediction results of proposed and baseline methods for the target session, S5-3. The red line is the predicted DI by the proposed method. The blue line is predicted by [9]. The gray line is predicted by the random selection method. The black line is the ground truth, the DI derived from the subject's behavioral responses. The correlations between the actual and predicted DIs are also listed. The proposed method selected the subjects who have similar EEG characteristics for transfer learning, so that the trained model was able to recognize drowsy episodes of the new subject.CC-BY-NC-ND · From the PDF

Verification

Claims the paper makes

The claims and statistics of this paper haven’t been checked yet.

Free: it reads the open-access text and checks the first 5 claims that cite a source we can check. Create a free account to check every claim.

A few free checks a day · papers already verified open instantly · results are public

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?

Self-check

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

The self-check needs the paper’s full text.

Citations

Cites

Show all 54 references

Cited by

See all 38

Metadata

Authors
Jeng, P.-Y., Wei, C.-S., Jung, T.-P., & Wang, L.-C.
Affiliations
  • National Yang Ming Chiao Tung University (TW)
  • University of California San Diego (US)
Venue
IEEE Journal of Biomedical and Health Informatics
Published
2021-06
Volume
25
Issue
6
Pages
1915-1925
ISSN
  • 2168-2194 (ISSN-L)
  • 2168-2208
OpenAlex
W3088964528
Cited by
35
References
53
Citations by year
2026: 2 · 2025: 6 · 2024: 8 · 2023: 8 · 2022: 5 · 2021: 6
FWCI
2.56
Type
journal article
Language
en
Open access
Locations
  • IEEE Journal of Biomedical and Health Informatics (publishedVersion, cc-by-nc-nd)
  • PubMed (publishedVersion)
Funders
  • Ministry of Education (108DR012)
  • National Chiao Tung University
  • Ministry of Science and Technology, Taiwan (107-2917-I-009-008, 108-2634-F-009-006, 109-2222-E-009-006-MY3)
  • Army Research Laboratory (W911NF-10-2-0022, W911NF2020088)
  • US National Science Foundation (CBET-1935860)
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
EEG decoding, transfer learning, passive brain-computer interface, unsupervised calibration, tensor decomposition, drowsiness detection, interindividual variability, neuromonitoring
MeSH terms
Machine Learning, Algorithms, Brain, Electroencephalography, Humans, 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.

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: Low dimensional subject representation based transfer learning in eeg decoding. https://app.cochat.ai/papers/10.1109/jbhi.2020.3025865
The paper
Jeng, P.-Y., Wei, C.-S., Jung, T.-P., & Wang, L.-C. (2021). Low dimensional subject representation based transfer learning in eeg decoding. IEEE Journal of Biomedical and Health Informatics. https://doi.org/10.1109/jbhi.2020.3025865

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.