-
Notifications
You must be signed in to change notification settings - Fork 104
Support ICA on complex-valued data #1056
Copy link
Copy link
Closed as not planned
Labels
decompositionissues related to decomposition methodsissues related to decomposition methodsdiscussionissues that still need to be discussedissues that still need to be discussedeffort: highMore than 40h total workMore than 40h total workenhancementissues describing possible enhancements to the projectissues describing possible enhancements to the projectimpact: mediumImproves code/documentation functionality for some usersImproves code/documentation functionality for some users
Description
Metadata
Metadata
Assignees
Labels
decompositionissues related to decomposition methodsissues related to decomposition methodsdiscussionissues that still need to be discussedissues that still need to be discussedeffort: highMore than 40h total workMore than 40h total workenhancementissues describing possible enhancements to the projectissues describing possible enhancements to the projectimpact: mediumImproves code/documentation functionality for some usersImproves code/documentation functionality for some users
Summary
I think we will start seeing more phase data from multi-echo protocols in the future.
Running ICA on complex-valued data can apparently produce better components than magnitude-only ICA (e.g., Yu et al., 2015).
Additional Detail
This is distinct from #1022, which would apply complex ICA to magnitude-only data.
I think GIFT has an implementation of entropy bound minimization (EBM) ICA, which apparently works better than FastICA on complex-valued data. We could maybe translate it to Python.