Data Modeling and Analytics
Rigorous data gathering and analysis methods
Many sources—including sensors, log files, databases, and qualitative means—are producing data at previously unimaginable scales and in unprecedented detail. The volume of data has outpaced the existing ability, in government and elsewhere, to use that data to make decisions, enable system adaptation, and generate input to train machine learning (ML) algorithms.
We apply cutting-edge techniques from academic and commercial data analytics innovation, often through collaboration with experts at Carnegie Mellon University. We develop ML algorithms and train them by mining the ground-truth data sets that we curate as an FFRDC. In that way, we tailor those technologies for the benefit of the Department of Defense, supporting our work in software cost estimation and control, human-machine teaming, autonomy and counter-autonomy, and system verification and validation.
We offer an approach that reduces risk and simplifies the selection and acquisition of big data technologies when you acquire and develop big data systems.
Costs for large new systems are hard to estimate. We developed a method to quantify uncertainty and increase confidence in a program's cost estimate.
August 17, 2016 • Technical Note
This technical note introduces Quantifying Uncertainty in Early Lifecycle Cost Estimation (QUELCE), a method for estimating program costs early in development.Download
May 16, 2016 • Conference Paper
John KleinRoss Buglak (Data to Decisions Cooperative Research Centre)David Blockow (Data to Decisions Cooperative Research Centre)Troy Wuttke (Data to Decisions Cooperative Research Centre)Brenton Cooper (Data to Decisions Cooperative Research Centre)
This paper presents a reference architecture for big data systems that is focused on addressing typical national defense requirements and that is vendor-neutral.Download
December 09, 2015 • Conference Paper
Eric Holk (Indiana University)Scott McMillanAndrew Lumsdaine (Indiana University)Jonathan ChuJohn MattySamantha MisurdaMarcin Zalewski (Indiana University)Peter Zhang (Indiana University)
Presented at the 2015 Supercomputing Conference, this paper shows that dynamic parallelism enables relatively high-performance graph algorithms for GPUs.Download
November 18, 2015 • Presentation
This presentation includes a brief demonstration of tools created by SEI staff that help scan, analyze, and prepare data to be used on a weekly metrics report.Download