Towards best practice of interpreting deep learning models for EEG-based brain computer interfaces
IntroductionAs deep learning has achieved state-of-the-art performance for many tasks of EEG-based BCI, many efforts have been made in recent years trying to understand what have been learned by the models. This is commonly done by generating a heatmap indicating to which extent each pixel of the in...
| Main Authors: | , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Frontiers Media S.A.
2023-08-01
|
| Series: | Frontiers in Computational Neuroscience |
| Subjects: | |
| Online Access: | https://www.frontiersin.org/articles/10.3389/fncom.2023.1232925/full |
| _version_ | 1828717303919804416 |
|---|---|
| author | Jian Cui Liqiang Yuan Zhaoxiang Wang Ruilin Li Tianzi Jiang Tianzi Jiang |
| author_facet | Jian Cui Liqiang Yuan Zhaoxiang Wang Ruilin Li Tianzi Jiang Tianzi Jiang |
| author_sort | Jian Cui |
| collection | DOAJ |
| description | IntroductionAs deep learning has achieved state-of-the-art performance for many tasks of EEG-based BCI, many efforts have been made in recent years trying to understand what have been learned by the models. This is commonly done by generating a heatmap indicating to which extent each pixel of the input contributes to the final classification for a trained model. Despite the wide use, it is not yet understood to which extent the obtained interpretation results can be trusted and how accurate they can reflect the model decisions.MethodsWe conduct studies to quantitatively evaluate seven different deep interpretation techniques across different models and datasets for EEG-based BCI.ResultsThe results reveal the importance of selecting a proper interpretation technique as the initial step. In addition, we also find that the quality of the interpretation results is inconsistent for individual samples despite when a method with an overall good performance is used. Many factors, including model structure and dataset types, could potentially affect the quality of the interpretation results.DiscussionBased on the observations, we propose a set of procedures that allow the interpretation results to be presented in an understandable and trusted way. We illustrate the usefulness of our method for EEG-based BCI with instances selected from different scenarios. |
| first_indexed | 2024-03-12T14:25:08Z |
| format | Article |
| id | doaj.art-4c89c4892cdf449db86f4639579b3564 |
| institution | Directory Open Access Journal |
| issn | 1662-5188 |
| language | English |
| last_indexed | 2024-03-12T14:25:08Z |
| publishDate | 2023-08-01 |
| publisher | Frontiers Media S.A. |
| record_format | Article |
| series | Frontiers in Computational Neuroscience |
| spelling | doaj.art-4c89c4892cdf449db86f4639579b35642023-08-18T07:40:27ZengFrontiers Media S.A.Frontiers in Computational Neuroscience1662-51882023-08-011710.3389/fncom.2023.12329251232925Towards best practice of interpreting deep learning models for EEG-based brain computer interfacesJian Cui0Liqiang Yuan1Zhaoxiang Wang2Ruilin Li3Tianzi Jiang4Tianzi Jiang5Research Center for Augmented Intelligence, Research Institute of Artificial Intelligence, Zhejiang Lab, Hangzhou, ChinaSchool of Electrical and Electronic Engineering, Nanyang Technological University, Singapore, SingaporeResearch Center for Augmented Intelligence, Research Institute of Artificial Intelligence, Zhejiang Lab, Hangzhou, ChinaSchool of Electrical and Electronic Engineering, Nanyang Technological University, Singapore, SingaporeResearch Center for Augmented Intelligence, Research Institute of Artificial Intelligence, Zhejiang Lab, Hangzhou, ChinaBrainnetome Center, Institute of Automation, Chinese Academy of Sciences, Beijing, ChinaIntroductionAs deep learning has achieved state-of-the-art performance for many tasks of EEG-based BCI, many efforts have been made in recent years trying to understand what have been learned by the models. This is commonly done by generating a heatmap indicating to which extent each pixel of the input contributes to the final classification for a trained model. Despite the wide use, it is not yet understood to which extent the obtained interpretation results can be trusted and how accurate they can reflect the model decisions.MethodsWe conduct studies to quantitatively evaluate seven different deep interpretation techniques across different models and datasets for EEG-based BCI.ResultsThe results reveal the importance of selecting a proper interpretation technique as the initial step. In addition, we also find that the quality of the interpretation results is inconsistent for individual samples despite when a method with an overall good performance is used. Many factors, including model structure and dataset types, could potentially affect the quality of the interpretation results.DiscussionBased on the observations, we propose a set of procedures that allow the interpretation results to be presented in an understandable and trusted way. We illustrate the usefulness of our method for EEG-based BCI with instances selected from different scenarios.https://www.frontiersin.org/articles/10.3389/fncom.2023.1232925/fullbrain-computer interface (BCI)convolutional neural networkdeep learning interpretabilityelectroencephalography (EEG)layer-wise relevance propagation (LRP) |
| spellingShingle | Jian Cui Liqiang Yuan Zhaoxiang Wang Ruilin Li Tianzi Jiang Tianzi Jiang Towards best practice of interpreting deep learning models for EEG-based brain computer interfaces Frontiers in Computational Neuroscience brain-computer interface (BCI) convolutional neural network deep learning interpretability electroencephalography (EEG) layer-wise relevance propagation (LRP) |
| title | Towards best practice of interpreting deep learning models for EEG-based brain computer interfaces |
| title_full | Towards best practice of interpreting deep learning models for EEG-based brain computer interfaces |
| title_fullStr | Towards best practice of interpreting deep learning models for EEG-based brain computer interfaces |
| title_full_unstemmed | Towards best practice of interpreting deep learning models for EEG-based brain computer interfaces |
| title_short | Towards best practice of interpreting deep learning models for EEG-based brain computer interfaces |
| title_sort | towards best practice of interpreting deep learning models for eeg based brain computer interfaces |
| topic | brain-computer interface (BCI) convolutional neural network deep learning interpretability electroencephalography (EEG) layer-wise relevance propagation (LRP) |
| url | https://www.frontiersin.org/articles/10.3389/fncom.2023.1232925/full |
| work_keys_str_mv | AT jiancui towardsbestpracticeofinterpretingdeeplearningmodelsforeegbasedbraincomputerinterfaces AT liqiangyuan towardsbestpracticeofinterpretingdeeplearningmodelsforeegbasedbraincomputerinterfaces AT zhaoxiangwang towardsbestpracticeofinterpretingdeeplearningmodelsforeegbasedbraincomputerinterfaces AT ruilinli towardsbestpracticeofinterpretingdeeplearningmodelsforeegbasedbraincomputerinterfaces AT tianzijiang towardsbestpracticeofinterpretingdeeplearningmodelsforeegbasedbraincomputerinterfaces AT tianzijiang towardsbestpracticeofinterpretingdeeplearningmodelsforeegbasedbraincomputerinterfaces |