Calendar of Events
Using Neural Networks to Select Galaxy Samples with Highly Accurate Photometric Redshifts for Cosmological Analysis
Speaker: Adam Broussard (Rutgers University)
The Legacy Survey of Space and Time will observe 25 billion galaxies, generating a key question: given this surfeit of statistics but only six photometric bands, can we select a “Platinum Sample” of galaxies that have highly accurate redshift (and therefore distance) estimates to better probe large-scale structure? We have performed redshift fits on Subaru Strategic Program galaxies using a random forest implementation known as Trees for Photo-Z (TPZ) with the goal of identifying galaxies with high-accuracy photometric redshifts. Because samples of galaxies with known redshifts are typically brighter and redder than those for which we estimate redshifts, we approximate the fit accuracy for dim galaxies by generating training and test sets with comparable differences in feature space. We find that training a neural network classifier to identify galaxies with ? < 0.07 (1+z) using galaxy photometry and associated TPZ fits improves the standard deviation of the error distribution by a factor of several while keeping roughly a fifth of the sample.