Software Engineering Institute

Machine learning (ML) is becoming a primary mechanism for extracting information from data. Indeed, each year, academic and research organizations publish more than 1,000 technical papers on ML, few (if any) of which provide enough detail to reproduce the results. The state of reporting on ML research slows adoption by DoD of advances needed to effectively use the surging volume of Big Data from network logs, internet activities, and sensory advancements. In this work, we are building a performance measurement workbench with tools to provide sound, reproducible ways to measure and report performance of large-scale ML platforms designed to operate on Big Data workloads.