Project submission: Unbiased electrode selection

[snastase]

Group name: Unnatural Selection
Group members: Sam

Project description: This is a "boring" project, but tackles an important part of our pipeline with ramifications for many projects. In many of our existing publications, we use one model (typically a model of low interest, like GloVe) to select "good" electrodes; then we run our analyses using the model of interest (e.g., GPT-2) in that subset of electrodes. What makes a "good" electrode? Presumably we only want to analyze electrodes that have some role in speech or language processing... But how do we determine this in a way that isn't circular (Kriegeskorte et al., 2009)? —in a way that doesn't bias our final results?

In this project, we'll aim to do the following:
(1) Build a tutorial-style Jupyter Notebook to demo our methods
(2) Develop one or more new methods for electrode selection
(3) Validate these methods on real data and simulated data
(4) Wrap these methods up into a class for sklearn.feature_selection