Question with asking how different or similar are three or more conditions to
each other?
Then we have a similarity matrix. For each new example, how similar are the
new examples to the old examples of all classes can be asked.
We can measure distance to the clouds of other classes to understand the
similarity that we observe. We can have information in non-diagonal cells of
similarity matrix to understand more interesting information about relation
between these classes.
Classification and Generalization in general can be encompassed by RSA.
We can have theoretical R.Similarity Matrices and observe region in RDMs
from experiments to see output similarity to original theoretical idea.
Comparing monkey IT region to Human FMRI data. – Research from Kriegeskorte
Many regions can separate images as classification but we can look at
off-diagonal cells and understand what property is helping make that
classification in that region.
You have a RSM. To know which model does this data fit to [Theoretical Bases],
we are faced with a model selection problem.
The critical power of RSA is you can have one good collected data and now you
can build more intelligent models later on to see if they explain the data
more than your previous models.
Behavioral analysis theory, Binary Matrices data, Continuous data, Other
modalities of neuro-imaging can be used as different models to compete with.
Noise Ceiling: Using the same conditions and collecting data again and then
making a matrix of how good your data quality is.