Here we will load data from Bonner and Epstein, 2017 . The data consist of fMRI responses to indoor scenes from scene selective regions (also whole brain) and navigational affordance behavioral responses in the form of representational dissimilarity matrices (RDMs) .
RSA is a method to relate signals from different source spaces (such as behavior, neural responses, DNN activations) by abstracting signals from separate source spaces into a common similarity space. For this, in each source space, condition-specific responses are compared to each other for dissimilarity (e.g., by calculating Euclidean distances between signals), and the values are aggregated in so-called representational dissimilarity matrices (RDMs) indexed in rows and columns by the conditions compared. RDMs thus summarize the representational geometry of the source space signals. Different from source space signals themselves, RDMs from different sources spaces are directly comparable to each other for similarity and thus can relate signals from different spaces
The figure below illustrates how RSA can be applied to different problems by comparing RDMs of different modalities/species.
There are 50 images of indoor scene in this dataset. These images were used to collect fMRI and behavioral responses to investigate coding of navigational affordances in the human visual system.