EEG Classification with Limited Data: A Deep Clustering Approach

Emotion recognition BC2214

Cognitive load detection SQ7322

PyTorch implementation

This repository contains a PyTorch implementation that simulates EEG-like signals using CIFAR-10 images and performs classification with:

• Feature generator (φ) using 1D convolutions + GRU + normalization
• Classifier head (C)
• Dictionary-based clustering head (P)
• Domain discriminator (D) for domain/adversarial simulation

Paper Reference

This project is based on the following paper:

Tabejamaat, M., Mohammadzade, H., Negin, F., & Bremond, F. (2025). EEG classification with limited data: A deep clustering approach. Pattern Recognition, 157, 110934. Link

BibTeX citation:

@article{tabejamaat2025eeg,
  title={EEG classification with limited data: A deep clustering approach},
  author={Tabejamaat, Mohsen and Mohammadzade, Hoda and Negin, Farhood and Bremond, Francois},
  journal={Pattern Recognition},
  volume={157},
  pages={110934},
  year={2025},
  publisher={Elsevier}
}

Folder Structure

EEG-DeepClustering/
├─ eeg_clustering/
│   ├─ __init__.py
│   ├─ dataset.py
│   ├─ model.py
│   ├─ train.py
│   └─ utils.py
├─ images/            
├─ notebooks/       
├─ tests/             
├─ data/              
├─ requirements.txt
├─ README.md
└─ LICENSE

Installation

Clone the repository:

git clone https://github.com/jedit-bit/EEG-DeepClustering.git
cd EEG-DeepClustering

Install dependencies:

pip install -r requirements.txt

Dependencies:

• torch
• torchvision

Usage

Run the training and evaluation script:

python eeg_clustering/train.py

• Trains the feature generator, classifier, clustering head, and domain discriminator.
• Uses CIFAR-10 images as EEG-like sequences (resized to 16x16).
• Prints training loss and accuracy for each epoch.
• Evaluates on the test set and prints final test accuracy.

Notes

• This is a non-official implementation inspired by EEG signals applied to CIFAR-10.
• Dictionary features are updated every epoch from one representative per class.
• Domain discriminator simulates multiple “subjects” or domains for adversarial training.

License

This project is released under the MIT License. See LICENSE for details.