Hi AFNI team,
I am using 3dsvm for MVPA analysis of task-based fMRI data and had a question about interpretation of the FIM outputs. My understanding is that, for a linear SVM, these maps represent the classifier weight vector used for pattern discrimination. Could they be treated as true “activation maps” in the same sense as univariate GLM activation maps or are the FIM outputs more like discriminative weight maps than direct measures of neural activation. I have also seen recent MVPA papers discuss weight-transformation methods (e.g., Haufe transformations) to convert classifier weights into more interpretable activation patterns.
So, essentially, my main questions are:
- How should the 3dsvm FIM outputs ideally be interpreted?
- Is it appropriate to refer to them as “activation maps,” or is it more accurate to describe them as classifier weight maps / discriminative pattern maps?
- Are there recommended approaches for making these maps more interpretable at the group level?
Thanks very much,
Valerie