Classification, Reconstruction and RSA

Question 1

What are the rows and columns of the data matrix in MVPA corresponding to?

Answer 1

rows = time points
columns = voxels

Question 2

What are the 3 spatial variants of multivariate classification?

Answer 2

• whole-brain
• region of interest
• searchlight

Question 3

What are the temporal variants of multivariate classification?

Answer 3

• trial-wise BOLD Signal
• run-wise GLM parameters

Question 4

What is a sample, feature and feature vector in multivariate classification?

Answer 4

• sample: specific brain data at one time point (one instance of a condition)
• feature/dimension: (activity of) voxel1, voxel2…
• feature vector: all features for one sample

Question 5

When does a classifier perform perfectly on training data?

Answer 5

when there are more features than samples in the training data

Question 6

What is the linear read-out hypothesis?

Answer 6

linear methods access what is explicitly encoded in the brain

Question 7

representational similarity analysis

Answer 7

• idea: do not look at (fMRI) patterns themselves, but at structure of pairwise (dis-)similarities
• comparing representation dissimilarity matrices of different conditions, models (with data), modalities, areas, subjects, species

Question 8

visual image reconstruction

Answer 8

-optimal: full mapping how a high-dimensional fMRI response r depends on a high-dimensional visual stimulus s
- reconstruction of stimulus from response
- full encoding model not accesible in practice
- decoding model: reconstruct single stimulus element
- encoding model: predict single response element

Question 9

Connectivity analysis

Answer 9

• typically on pairs of uni-variate time courses, not on multi-variate patterns
• how are signals at two different locations related?