ASPEN: Spectral-Temporal Fusion for Cross-Subject Brain Decoding

This framework, Adaptive SPectral Encoder Network (ASPEN), aims to improve the cross subject generalization in EEG signals.

Data Preprocessing

The data used in our analysis and benchmarking was from MOABB.

After downloading the datafiles, to preprocess the data:

  cd data_process
  python preprocess_data.py

ASPEN

To train (options for task name are 'SSVEP', 'Lee2019_SSVEP', 'BI2014b_P300', 'BNCI2014_P300', 'MI', 'Lee2019_MI'):

  cd model
  python -m train_aspen —task [task name]

SPEN (SPectral Encoder Network)

SPEN is only the spectral stream of the network. This doesn't incorporate the temporal signals.

To train (options for task name are 'SSVEP', 'Lee2019_SSVEP', 'BI2014b_P300', 'BNCI2014_P300', 'MI', 'Lee2019_MI'):

  cd model
  python -m train_spen —task [task name]

To test (options for task name are 'SSVEP', 'Lee2019_SSVEP', 'BI2014b_P300', 'BNCI2014_P300', 'MI', 'Lee2019_MI'):

  python -m test_spen —task [task name]

Acknowledgements

This work was done as a part of the CMU 11-785: Introduction to Deep Learning course.

References

Baselines

• EEGNet
• EEGConformer
• TSformer-SA
• MultiDiffNet
• CTNet

Datasets

• MOABB