How to train a neural network, working on eeg, to deal with missing channels

I am building a data set which will be used to train a neural network to recognize specific signals in 20 channels of eeg (i.e. input is 20xN values). Some of the data used to train this network (as well as expected future input) will only contain 18 or 19 of these channels. Is there a way to train this network to deal with these unknowns????

(I have seen articles on averaging values and the like. I'm looking for something a bit more sophisticated?)

Answer 1

[Greg Heath]

You may have to create a series of increasingly sophisticated interpolation techniques.

1. Start with a baseline of zeros for missing data 2. Next try linear interpolation 3. Typically, that's it for me. However, you could move on to quadratic interpolation or fourier series.

Hope this helps.

 *Thank you for formally accepting my answer*

Greg