Fabian Gieseke (Carl von Ossietzky Universität Oldenburg), Kai Lars Polsterer (Ruhr-Universität Bochum), Peter Christian Zinn (Ruhr-Universität Bochum), Oliver Kramer (Carl von Ossietzky Universität Oldenburg)
The task of estimating an object's redshift based on photometric data is one of the most important ones in astronomy. This is especially the case for the redshift estimation of quasi-stellar radio sources (quasars), which will be subject of the work at hand. Common approaches for this regression task in the field of astronomy are often based on nearest neighbor search or template fitting schemes. However, due to lack of labeled data for distant quasars, such methods often yield unsatisfying results caused by the sparsity in the corresponding regions of the feature space. A wide range of regression techniques exists in the field of machine learning. Among the most popular ones are regularized regression schemes like ridge regression or support vector regression. Due to both their empirical performance on real-world data as well as their theoretical properties in the context of statistical learning theory, such schemes have gained a considerable interest in various application fields. Surprisingly, methods of this type have not yet been investigated extensively for the task of estimating redshift estimation of quasars. The purpose of this work is to analyze the empirical performance of such methods for this task. The experiments are conducted based on both extracted features like PSF magnitudes and raw images, which are available for all five bands in the Sloan Digital Sky Survey. In addition, some challenges and opportunities of recent advances in the field of machine learning including, e.g., missing data are discussed.
Paper ID: P049