Deep learning strategies for identifying and characterizing EM ducts within the marine atmospheric boundary layer
Abstract
The goal of the currently proposed research is to apply state of the artdeep learning methods (e.g. convolutional neural networks (CVNs), transferlearning CVNs, and deep neural networks) to the data-driven MABL ductheight inversion problem. Preliminary work done in the PI s group indi-cates that these approaches are immensely promising for inferring MABLduct heights: o -ering speed, accuracy, precision, and robustness to noise;though this application does a -ord fundamental challenge for machine learn-ing approaches.In the world of engineering and applied science, data are expensive toacquire, and as a result, these data are usually sparse in nature (money isconserved by measuring as little as possible), biased (we usually measurewhen and where we can: where it is convenient and safe), and noisy (i.e.containing electronic sensor noise, possessing spatio-temporal sensor loca-tion uncertainty, having calibrations that are o -, or are not stable, etc.)The present research will develop novel strategies for dealing with these dif-?culties. These strategies will almost surely be of general interest to otherapplications where ML is desired for application within the real world.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 24, 2019
- Source ID
- N000141912095
Entities
People
- Christopher Earls
Organizations
- Cornell University
- Office of Naval Research
- United States Navy