Representational similarity analysis¶
Visualizing dissimilarity matrices¶
Dissimilarities, based on neural data, behavioural data, and/or model data, can be visualised in various ways.
Compute the similarities in the
vt regions for participant
s01 across the six categories, and load the model similarities for the behavioural ratings. Then visualize the similarities in three ways:
with dissimilarity matrices
with multi-dimensional scaling
Hint: run rsa visualize skl.
Comparing dissimilarity matrices¶
It is easy to compare dissimilarity matrices by computing the Pearson correlation between two flattened upper triangle DSMs using the cosmo corr function. For the next exercise, stack flattened DSMs vertically into a single matrix starting with all of the EV DSMs from every subject then all of the VT DSM. You should have an 16x15 matrix. Then add the v1 model and the behavioral DSMs to make it a 18x15 matrix. Now compute the cross-correlation matrix using cosmo corr. Visualize the cross-correlation matrix with imagesc. Try this with demeaning and without demeaning to compare the results. Finally, use matlabs boxplot function to view the distributions of correlations between neural simiilarities and model/behavioral DSMs.
Hint: run compare dsm skl.
Target dissimilarity matrix searchlight¶
The function cosmo target dsm corr measure implements representational similarity. Use this measure to map where the neural similarity is similar to the behavioural similarity.
It is recommended to center the data using the
Advanced exercise: the cosmo target dsm corr measure function can also run regression on multiple dissimilarity matrices. Use this function to estimate the contribution of the V1 and behavioural model using a searchlight.