In my last blog post, I described how Wasserstein (a.k.a. Earth-Mover’s) distances could be used to measure the dissimilarity between two neural response patterns. The main benefit of “Brain-Mover’s Distance” is that it takes the topology of the brain into account, measuring not just how similarly the voxels in question respond, but also their proximity in the brain. I also worked through an example to demonstrate how this method could be used to assess the replicability or inter-subject reliability of an fMRI dataset.
When analyzing fMRI data, we often seek to measure similarity between two brain responses. For example, we run reliability analyses to ask, “how similar is this subject’s brain when they see the same image again?” or “how similar is this subject to the rest of the group?” Relatedly, Representational Similarity Analyses (RSA) allow us to ask, “Does this brain region respond the same way to all members of a category - like inanimate objects?
How are action representations organized in the visual cortex?
How do we perceive how much effort other people are exerting?