ENSEMBLE LEARNING FOR INDEPENDENT COMPONENT ANALYSIS OF NORMAL GALAXY SPECTRA

In this paper, we employ a new statistical analysis technique, ensemble learning for independent component analysis (EL-ICA), on the synthetic galaxy spectra from a newly released high-resolution evolutionary model by Bruzual & Charlot. We find that EL-ICA can sufficiently compress the synthetic galaxy spectral library to six nonnegative independent components (ICs), which are good templates for modeling huge amounts of normal galaxy spectra, such as the galaxy spectra in the Sloan Digital Sky Survey (SDSS). Important spectral parameters, such as starlight reddening, stellar velocity dispersion, stellar mass, and star formation histories, can be given simultaneously by the fit. Extensive tests show that the fit and the derived parameters are reliable for galaxy spectra with the typical quality of the SDSS.