Representational Similarity Analysis
A core analytical tool at the intersection of deep learning and cognitive neuroscience, used to compare the internal representation structures of different models or neural systems. The field encompasses multiple specific metrics, including RSA (Representational Similarity Analysis), CKA (Centered Kernel Alignment), and Procrustes analysis.
Timeline (last 90 days)
A study proposes that differences between representational similarity metrics are far less important than the differences introduced by the objects of comparison (i.e., how data is organized).
With the number of categories held constant, changing which categories are compared shifts the distance between two representations by more than twice the measured effect.
Changing how many categories are compared has almost no impact on representational similarity conclusions.
Switching to a better metric cannot salvage an analysis that already contains arbitrariness at the data construction stage.
All Facts (4)
A study proposes that differences between representational similarity metrics are far less important than the differences introduced by the objects of comparison (i.e., how data is organized).
50%UnverifiedWith the number of categories held constant, changing which categories are compared shifts the distance between two representations by more than twice the measured effect.
50%UnverifiedChanging how many categories are compared has almost no impact on representational similarity conclusions.
50%UnverifiedSwitching to a better metric cannot salvage an analysis that already contains arbitrariness at the data construction stage.
50%