Testbed for Quantifying Uncertainty in Ocean Acoustics Machine Learning

Abstract

Because measurements at sea are expensive, we propose a laboratory testbed to explore how the generalization error changes as the m"easurement environment varies from the environment used in the training data set. The testbed will be equipped for variable bathymetry and sediment layers, with possible range and azimuthal dependence. The testbed will be designed to also handle both distilled and seawater to allow for a vertically stratified sound speed profile, as well as a time-varying one. ML algorithms trained and tested on data simulated for one testbed configuration, using established range-independent or range-dependent propagation algorithms, will then be applied to measurements from the testbed for different configurations. The numerous testbed configuration will allow for the impact of different environmental features on the generalization error to be studied independently and jointly. In addition, the testbed will also be used to simulate noisy ocean environments, in which a target~s signal may be masked by interferers, by adding additional sources. A computer-controlled positioning system will be employed to simulated source motion and measurement along a mu"ltiple-hydrophone array~ providing a controlled method for quantifying the effect of those events on the ML algorithm performance.

Document Details

Document Type
DoD Grant Award
Publication Date
Aug 20, 2019
Source ID
N000141912683

Entities

People

  • Tracianne B Neilsen

Organizations

  • Brigham Young University
  • Office of Naval Research
  • United States Navy

Tags

Readers

  • Acoustical Oceanography.
  • Computer Networking
  • Neural Network Machine Learning.

Technology Areas

  • AI & ML
  • AI & ML - Autonomous Systems
  • AI & ML - Bayesian Inference
  • AI & ML - Neural Networks