1. Aung Zar Lin (st121956)
2. Lin Tun Naing (st122403)
3. Min Khant Soe (st122277)
4. Win Win Phyo (st122314)

0. Visualize ERP
1. Load_data_pool.ipynb
2. Load_data
3. BN3 subject specific
4. BN3 train on one subject and test on another subject
5. BN3 regional channel
6. BN3+LSTM subject specific
7. BN3+LSTM train on one subject and test on another subject
8. BN3+LSTM regional channel
9. CNN subject specific
10. CNN train on one subject and test on another subject
11. CNN regional channel
12. BN3 regional channel P Region Test
13. BN3+LSTM regional channel P Region Test
14. CNN regional channel P Region Test
15. BN3 subject pool
16. BN3+LSTM subject pool
17. CNN subject pool
18. LDA train on one subject and test on another subject
19. SVM train on one subject and test on another subject

-- 'data' Folder contains '_epo.fif' files

We loaded all data from .csv files and turned into "_epo.fif" formats. Each coding process includes in "Load_data.ipynb" file. The following preprocessing steps are done.

1. Load csv
2. Convert to raw - set montage, eeg channels, targets and sampling frequency
3. Filter - electrical band, -0.1 to 0.7 band pass
4. Epoching - epoch "Target" and "Non-target"
5. save - as "_epo.fif" format

"Load_data_pool.ipynb" loads all the "_epo.fif" files and convert each and every of them into numpy arrays and concatenate each others into X and y. The sample shape becomes (34236, 32, 410) for BN3 Conv1D or (34236, 1, 32, 410) for CNN Conv2D.

We used total of 3 models named BN3 (Batchnormalization Conv1D), BN3+LSTM and CNN Conv2D.

The experiment includes the following.

1. Trained and test on one subject and test on another subject.
     - Classical machine learning models: SVM and LDA as baselines
     - BN3, BN3+LSTM, CNN Conv2D
2. Subject specific
     - BN3, BN3+LSTM, CNN Conv2D
3. Regional channel
     - BN3, BN3+LSTM, CNN Conv2D
4. Subject Pool
     - BN3, BN3+LSTM, CNN Conv2D

• Data are saved as (.fif) file in folder in order to save time when we want to run the data in model again.
• Build CNN model and recorded the accuracy in excels for both 1 subject and each channels of that subject.
• Compared accuracy result with ML models.

Expected to finish next week

• Improve CNN model as possible as to beat the accuracy of ML
• Build LSTM+CNN model

modeling - 80 % In this week, we do P300 analysis and continue working on classification models.

Expected to do next week -inter-brain synchrony

modeling - 50 % In this week, we fit the data with 4 models (ML, LSTM, CNN with 1d and 2d) and compare the accuracy.

Expecting to work on the next week

• p300 Analysis

preprocessing and modeling - 25%

We have done in this week.

• ICA (difficult to differentiate bettween eye,muscle artifacts and signals)
• Epoching

Expecting to work on the next week

• modeling

modeling - 7% loading the dataset - 100% creating the github - 100% reading paper - 90 %

we read the following paper

• MEG and EEG data analysis with MNE-Python (https://www.frontiersin.org/articles/10.3389/fnins.2013.00267/full)

• Universal Joint Feature Extraction for P300 EEG Classification Using Multi-Task Autoencoder (https://ieeexplore.ieee.org/abstract/document/8723080)

• Lawhern, Vernon J. EEGNet: A Compact Convolutional Network for EEG-based Brain-Computer Interfaces (https://www.researchgate.net/publication/310953136_EEGNet_A_Compact_Convolutional_Network_for_EEG-based_Brain-Computer_Interfaces)

• Asscement of preprocessing of classifiers on using P300 paradigm (https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=4462003)

• A P300 event-related potential brain–computer interface (BCI): The effects of matrix size and inter stimulus interval on performance (https://www.sciencedirect.com/science/article/pii/S0301051106001396)

We done this steps in data processing:

• transform data in to raw mne object
• notch filter
• band pass filter

• got error in notch filter coding and try to fix
• ICA
• epoching

task-based modeling - 0% loading the dataset - 100% creating the github - 100% reading paper - 40 %

we read the following paper

• Experimental procedures for this dataset (https://hal.archives-ouvertes.fr/hal-02173958/document?fbclid=IwAR2HuX-mjDmMsokUg2zYyPkHnI-WyfX0oIRWgKAffLgJ7yO0pbyM9mNY7Q8)

• A novel P300 BCI speller based on the Triple RSVP paradigm (https://www.nature.com/articles/s41598-018-21717-y)

• The Changing Face of P300 BCIs: A Comparison of Stimulus Changes in a P300 BCI Involving Faces, Emotion, and Movement (https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0049688)

• Progress in neural networks for EEG signal recognition in 2021 (https://arxiv.org/ftp/arxiv/papers/2103/2103.15755.pdf )

• Trends in EEG-BCI for daily-life: Requirements for artifact removal (https://www.sciencedirect.com/science/article/pii/S1746809416301318#bib0610)

• Analysis of P300 Related Target Choice in Oddball Paradigm (https://www.koreascience.or.kr/article/JAKO201120661418465.page)

• A Novel P300 Classification Algorithm Based on a Principal Component Analysis-Convolutional Neural Network (https://www.mdpi.com/2076-3417/10/4/1546/htm)