Project work 4 (1st-dl 9th February at 10:00)
The goal of this week’s project work is to perform representational similarity analysis (RSA) on the fMRI dataset.
How? I suggest you do the following, but feel free to play with the data and come up with novel ways, for example, to visualize the results:
(1) Extract the response patterns (e.g., t maps or beta maps) from different regions-of-interests for all stimulus categories
(2) Construct the RDMs, separately for each region-of-interest and for each subject.
- see, e.g., matlab function ‘pdist.m’ or use 'corr.m'
(3) Visualize RDMs—Which stimulus directions are emphasized? Which are de-emphasized?
- see, e.g., (a) ‘imagesc.m’ and (b) multidimensional scaling (‘mdscale.m’ with 'metricstress‘ as the criterion)
(4) How similar are the RDMs across regions-of-interest and between subjects?
(5) If you have the data divided to training & testing datasets (related to last week’s project), you can make the split-data RDMs to (1) study the replicability of the RDMs, and/or (2) calculate the exemplar discriminability indices (EDIs).
Return the first version of your project work by Tuesday, February 9, at 10:00. Even if you have had trouble with the analysis, do return something that we can then discuss during the Tuesday's Q&A session. You will get 5 points for returning your preliminary results and actively contributing during the Q&A session.