The choice of target channels

[Daniel-Lin-S]

In reviewing your code and methodology, I noticed that for some datasets—such as BCICIV—the input channels are mapped to a set of target channels using a 1D convolution layer with a kernel size of 1. For example, the 3 channels of BCICIV-2b are mapped to 7 channels (['C5', 'C3', 'C1', 'CZ', 'C2', 'C4', 'C6']), and the 22 channels of BCICIV-2a are mapped to 19 target channels.

I’m very interested in understanding the rationale behind these specific mappings. Could you kindly share more about how these target channels were chosen? Additionally, how do you ensure that the mapped signals remain compatible with the channel space used during pre-training?