Techniques for massive-data machine learning in astronomy

Abstract

Important computational algorithms for statistical analysis of massive datasets will require efficient NlogN implementations. A leading group producing these algorithms is the FASTlab group at Georgia Institute of Technology. Substantial speedups over naive algorithms are achieved; for example, from O(N3) to O(N) for Support Vector Machine classification and from O(Nn) to O(Nlogn) for npoint correlation functions. These methods can be applied to datasets such as the massive image dataset from the Next Generation Virgo Cluster Survey hosted at the Canadian Astronomy Data Centre. Object classification, Virgo Cluster membership, photometric redshifts, catalog cross-matching, and spatial clustering can potentially be achieved with greatly improved efficiency. © Springer Science+Business Media New York 2013.