We study the composition of the faint radio population selected from the VLA-COSMOS 3GHz Large Project, a radio continuum survey performed at 10 cm wavelength. The survey covers the full 2 square degree COSMOS field with mean $rms\sim2.3$ $\mu$Jy/beam, cataloging 10,899 source components above $5\times rms$. By combining these radio data with UltraVISTA, optical, near-infrared, and Spitzer/IRAC mid-infrared data, as well as X-ray data from the Chandra Legacy, and Chandra COSMOS surveys, we gain insight into the emission mechanisms within our radio sources out to redshifts of $z\sim5$. From these emission characteristics we classify our souces as star forming galaxies or AGN. Using their multi-wavelength properties we further separate the AGN into sub-samples dominated by radiatively efficient and inefficient AGN, often referred to as high- and low-excitation emission line AGN. We compare our method with other results based on fitting of the sources' spectral energy distributions using both galaxy and AGN spectral models, and those based on the infrared-radio correlation. We study the fractional contributions of these sub-populations down to radio flux levels of $\sim$10 $\mu$Jy. We find that at 3 GHz flux densities above $\sim$400 $\mu$Jy quiescent, red galaxies, consistent with the low-excitation radio AGN class constitute the dominant fraction. Below densities of $\sim$200 $\mu$Jy star-forming galaxies begin to constitute the largest fraction, followed by the low-excitation, and X-ray- and IR-identified high-excitation radio AGN.
7 pages, 3 figures, The many facets of extragalactic radio surveys: towards new scientific challenges, Bologna 20-23 October 2015