MUPS-EEG

Code for MUPS-EEG model introduced in "Ultra Efficient Transfer Learning with Meta Update for Cross Subject EEG Classification".

Setting up environment

The model is implemented with Pytorch, we recommend python 3.5 and PyTorch 0.4.0 with Anaconda.

Create a new environment and install python packages in it:

   conda create --name mups python=3.5
   conda activate mups
   conda install pytorch=0.4.0 -c pytorch
   conda install scipy scikit-learn

Install other dependencies:

   pip install six tqdm tensorboardX

Clone the repository:

   git clone https://github.com/tiehangd/MUPS-EEG

Dataset preparation

Download BCI-IV 2a dataset from http://bnci-horizon-2020.eu/database/data-sets, Four class motor imagery (001-2014)

Place the 18 files inside ./data folder

Data preprocess, run from command line

   python ./dataloader/data_preprocessing.py

This produces data for our cross subject task, which is stored in ./data/cross_sub

Running the model

1. Pretraining of feature extractor

   python pre_train.py
   

2. Meta adaptation

   python meta_adapt.py
   

Citation

Please cite our paper if it is helpful to your work:

@misc{duan2021ultra,
      title={Ultra Efficient Transfer Learning with Meta Update for Cross Subject EEG Classification}, 
      author={Tiehang Duan and Mihir Chauhan and Mohammad Abuzar Shaikh and Jun Chu and Sargur Srihari},
      year={2021},
      eprint={2003.06113},
      archivePrefix={arXiv},
      primaryClass={cs.LG}
}

Acknowledgements

Implementation of MUPS model utilized code from the following repositories:

1) https://github.com/yaoyao-liu/meta-transfer-learning
2) https://github.com/aliasvishnu/EEGNet