E1: Machine Learning 101 and PNT

Time:
Monday, June 2, 8:30 a.m. - 10:00 a.m.
Location: Ballroom E

This course introduces the fundamentals of machine learning (ML) and how it applies to position, navigation, and timing (PNT). Basic machine learning concepts like types of ML, importance and collection of data sets, deployment strategies, and the development toolchain will be covered. Some of the common pitfalls in ML developments will be highlighted along with strategies for avoiding them. This workshop will follow examples that encapsulate PNT expertise in algorithm form. Pursuit of the application of these algorithms in the OODA loop leads to demonstrations of how they might accelerate it, exposing the PNT OODA loop operational concept and enabling discussion of its architectural elements.

Training expert algorithms requires data, and the more varied and representative the dataset, the better their accuracy. While simulated data is often adequate for proof-of-concept neural networks, real-world data is necessary for accuracy in real-world environments. The tutorial will tackle questions of how, when and what to collect for the gathering of high-value datasets. Considerations for data-collection and specifications will be outlined, as well as ongoing challenges in combining datasets. Attendees will have the opportunity to experiment with tools developed for addressing unique challenges of applying ML to PNT.

This course employs a package of readily available ML tools that are either created or assembled for government use. These tools include a data collection system, a set of reference ML algorithms as a good starting point in that they have a good performance history in a contested environment, a ML development and test environment that employs standardized and compatible toolsets, and a set of vetted and conditioned data sets; everything that a ML developer or evaluator needs to get started on applying ML solutions for PNT in a contested environment.

Prerequisites: Basic understanding of optimization and the Python programming language are useful but not required for this course. Those that want to follow-along can bring a laptop, but it is optional.

Andrew Cochrane is an assistant research professor and Chief Engineer at the COSMIAC Research Center at the University of New Mexico's School of Engineering. His current work with the Air Force Research Laboratory is focusing on the development of ML tools for application to the PNT OODA loop. Dr. Cochrane brings a broad experience including signal design, signal processing, RF system testing, field experiment design, computer networks, embedded computing, system integration as well as a much deeper experience developing low-level physics software to the Kirtland AFB Satellite Navigation Technical Area (SatNavTA). His research interests include remote sensing, emerging signal types, hierarchical neural networks, EM-wave-material interactions and autonomous navigation.

Dr. James Aarestad is an assistant research professor and Chief Scientist at the COSMIAC Research Center at the University of New Mexico's School of Engineering. His current and prior work with the SatNavTA group within the Air Force Research Laboratory includes the implementation of solutions to the technical challenges faced by the SatNavTA team in the areas of hardware and software development, signal development research and the application of machine learning tools and hardware. Dr. Aarestad's research interests revolve around reconfigurable and programmable logic for signal processing and transmission.