Spectral Basis
A set of orthogonal basis vectors obtained through eigendecomposition (such as eigenvalue decomposition of a covariance matrix or parameter matrix), where eigenvalue magnitude reflects the information content or importance of the corresponding direction, in line with the idea of principal component analysis.
Timeline (last 90 days)
SpecMRL reorganizes the feature space via a parameter-derived spectral basis, achieving more effective prefix readout.
SpecMRL's spectral basis is derived from model parameters and is end-to-end learnable, rather than being a post-hoc decomposition of fixed representations.
A spectral basis is a set of orthogonal basis vectors obtained through eigendecomposition, where the magnitude of the eigenvalues reflects the information content or importance of the corresponding direction.
All Facts (3)
SpecMRL reorganizes the feature space via a parameter-derived spectral basis, achieving more effective prefix readout.
50%UnverifiedSpecMRL's spectral basis is derived from model parameters and is end-to-end learnable, rather than being a post-hoc decomposition of fixed representations.
50%UnverifiedA spectral basis is a set of orthogonal basis vectors obtained through eigendecomposition, where the magnitude of the eigenvalues reflects the information content or importance of the corresponding direction.
50%