Effect of auxiliary cross-paradigm data on paradigm-specific EEG foundation model pre-training

Determine whether incorporating EEG recordings from paradigms other than the target paradigm as auxiliary data during pre-training enhances the downstream performance of paradigm-specific EEG foundation models on that target paradigm.

Background

The paper compares general-purpose EEG foundation models with a paradigm-specific model (MIRepNet) on motor imagery tasks and finds that the paradigm-specific model achieves state-of-the-art results, supporting the value of task-focused pre-training.

Building on this observation, the authors argue that paradigm-specific pre-training is a promising direction for practical BCI deployment, but they note uncertainty about whether adding data from other paradigms during pre-training could further help or hinder performance. This raises a concrete open question about the role of cross-paradigm data in paradigm-specific model development.