Fred Reiss presented this deep dive at Spark Summit East in NYC in February 2016. See the slides on SlideShare
Inside Apache SystemML – Spark Summit East 2016 from Fred Reiss
Fred Reiss talks about the origins and early history of SystemML, including the team’s efforts to maintain the advantages of a small-data approach as data volume grows. In particular, he shows off code for handling alternating least squares using SystemML’s subset of R. By leveraging live variable analysis, propagation of statistics, and advanced rewrites of the R code, SystemML achieves sizable speed and efficiency gains.