Spectral Concentration
A metric based on eigenvalue analysis of the representation covariance matrix. It measures the proportion of total variance carried by a small number of leading eigenvalues.
Timeline (last 90 days)
The authors propose three asymmetric diagnostic metrics to distinguish linear world models from kernel machines: reverse predictiveness, spectral concentration, and out-of-distribution generalization.
Spectral concentration measures the degree of variance concentration through eigenvalue analysis of the feature covariance matrix, requires no labels, relies only on the features themselves, and can serve as an unsupervised diagnostic tool.
All Facts (2)
The authors propose three asymmetric diagnostic metrics to distinguish linear world models from kernel machines: reverse predictiveness, spectral concentration, and out-of-distribution generalization.
50%UnverifiedSpectral concentration measures the degree of variance concentration through eigenvalue analysis of the feature covariance matrix, requires no labels, relies only on the features themselves, and can serve as an unsupervised diagnostic tool.
50%