With large-scale galaxy surveys, we can observe hundreds of thousands of galaxies or more, up to billions with upcoming experiments such as WFIRST, Euclid and LSST. While such surveys cannot obtain spectra for all observed galaxies, we have access to the galaxy magnitudes in color filters. This data set behaves like a high-dimensional nonlinear surface, making it an excellent target for machine learning methods. In this work, we use a lightcone of semianalytic galaxies tuned to match CANDELS observations from Lu et al. (2014) to train a set of neural networks on a set of galaxy properties (stellar mass, metallicity, and average star formation rate) using the truth values from the semianalytic catalogs. We also demonstrate the effect of adding simulated observational noise to the simulated magnitudes, and then use neural networks trained on the noisy data to predict stellar masses, metallicities, and average star formation rates on real CANDELS galaxies, comparing our results to the physical parameters obtained from SED fitting. On semianalytic galaxies alone, we are nearly competitive with template-fitting methods. For the observed CANDELS data, our results are not as accurate, with indications that this inaccuracy is due to a combination of different assumptions in template-fitting and differences between the semianalytic models and the observed galaxies, particularly in the noise properties. Our results show that stellar mass, metallicity, and star formation rate can in principle be measured with neural networks at a competitive degree of accuracy and precision relative to physically motivated template-fitting methods if an appropriate training set can be obtained.

We investigate the clustering and halo properties of $\sim 5000$ Ly$\alpha$-selected emission line galaxies (LAEs) from the Slicing COSMOS 4K (SC4K) and from archival NB497 imaging of SA22 split in 15 discrete redshift slices between $z \sim 2.5 - 6$. We measure clustering lengths of $r_0 \sim 3 - 6\ h^{-1}$ Mpc and typical halo masses of $\sim 10^{11}$ M$_\odot$ for our narrowband-selected LAEs with typical $L_{\rm{Ly}\alpha} \sim 10^{42 - 43}$ erg s$^{-1}$. The intermediate band-selected LAEs are observed to have $r_0 \sim 3.5 - 15\ h^{-1}$ Mpc with typical halo masses of $\sim 10^{11 - 12}$ M$_\odot$ and typical $L_{\rm{Ly}\alpha} \sim 10^{43 - 43.6}$ erg s$^{-1}$. We find a strong, redshift-independent correlation between halo mass and Ly$\alpha$ luminosity normalized by the characteristic Ly$\alpha$ luminosity, $L^\star(z)$. The faintest LAEs ($L \sim 0.1\ L^\star(z)$) typically identified by deep narrowband surveys are found in $10^{10}$ M$_\odot$ halos and the brightest LAEs ($L \sim 7\ L^\star(z)$) are found in $\sim 5 \times 10^{12}$ M$_\odot$ halos. A dependency on the rest-frame 1500 \AA~UV luminosity, M$_\rm{UV}$, is also observed where the halo masses increase from $10^{11}$ to $10^{13}$ M$_\odot$ for M$_\rm{UV} \sim -19$ to $-23.5$ mag. Halo mass is also observed to increase from $10^{9.8}$ to $10^{12.3}$ M$_\odot$ for dust-corrected UV star formation rates from $\sim 0.6$ to $10$ M$_\odot$ yr$^{-1}$ and continues to increase up to $10^{13.5}$ M$_\odot$ in halo mass, where the majority of those sources are AGN. All the trends we observe are found to be redshift-independent. Our results reveal that LAEs are the likely progenitors of a wide range of galaxies depending on their luminosity, from dwarf-like, to Milky Way-type, to bright cluster galaxies. LAEs therefore provide unique insight into the early formation and evolution of the galaxies we observe in the local Universe.