Is Limited Participant Diversity Impeding EEG-based Machine Learning? (NeurIPS 2025)

This repository contains the research code for the experiments in our paper "Is Limited Participant Diversity Impeding EEG-based Machine Learning?", accepted at NeurIPS 2025.

Install

1. Create and activate conda/mamba environment

   mamba env create -f environment.yml
   mamba activate participant-diversity-paper
   

2. Install the package

   pip install -e .
   

3. Request access to the datasets and refer to the following repositories to harmonise and convert them to BIDS format.

   1. TUAB: https://github.com/bomatter/data-TUAB

   2. CAUEEG: https://github.com/bomatter/data-CAUEEG

   3. PhysioNet: https://github.com/bomatter/data-PhysioNet

4. Create a copy of the user_config.example.yml file and rename it to user_config.yml. Then open it and configure the paths to the dataset folders and the directory, where you want outputs to be saved.

5. To run experiments with the pretrained LaBraM model, download the labram-base.pth checkpoint to the root folder of this repository from the official labram repo.

Run Experiments

Individual trials can be run with the run_trial.py script.

python core/run_trial.py \
    --model TCN --dataset TUAB --task normality \
    --n_participants 10 --n_segments 10 \
    --output_root results/

Alternatively, use the run_experiments.ipynb notebook to generate configurations (and submit slurm jobs) for all experiments to reproduce the paper results.

Citation

@article{bomatter2025limited,
  title={Is Limited Participant Diversity Impeding EEG-based Machine Learning?},
  author={Bomatter, Philipp and Gouk, Henry},
  journal={arXiv preprint arXiv:2503.13497},
  year={2025}
}