Spectral theory of large data with power-law latent features
En palabras de los autores
Power-law eigenvalue spectra of covariance matrices estimated from data appear in diverse domains ranging from neural populations to proteins. We introduce a generative latent feature model for such data, optionally corrupted by measurement noise, and develop its spectral theory. We derive closed leading-order self-consistent equations for the Stieltjes transform of the covariance matrix from which the bulk eigenvalue density is obtained and provide a variational principle for the self-consistent equations. We also compute the positions and fluctuations of isolated top eigenvalues. The theory is validated in known algebraic limits and tested against simulations.
Apareció: viernes, 30 de octubre. arXiv. Preprint, todavía sin revisión por pares.
Comentario de los autores: 19 pages, 7 figures