Advancing Trustworthy Machine Learning for Seabed Morphodynamics Analysis
Abstract
Advances in AI and machine learning (ML), particularly deep learning (DL), have shown great promise in characterizing and understanding seafloor morphodynamics in support of naval applications. These applications often collect large-scale geoacoustic and environmental data across a wide spatiotemporal range . Using such data to gain scientific insights into complex phenomena requires characterizing dynamics and uncertainty among a large number of variables. This poses an unprecedented out-of-distribution (OOD) challenge where AI-ML models have to operate on unseen distributions that lie outside their training space, rendering sharp performance drops in prediction and explanation. (1) Fragile prediction- Despite the well-documented success, existing predictive AI-ML models are highly susceptible to distributional shifts. A predictive model that is accurate on average conditions can fail catastrophically when presented with rare or unseen distributions. (2) Unreliable explanation- Understanding the decision-making process is desirable for trustworthy modeling. However, existing explainable AI-ML models often rely on spurious correlations for prediction, yielding inconsistent explanations across distributions. To address these challenges, this project will develop a suite of new optimization methods aiming to advance the trustworthiness of AI-ML models on large distributed seafloor datasets. The research outcomes, including an AI-ready multi-site seabed morphodynamic database, innovative trustworthy AI-ML optimization methods, and scalable implementations of algorithms, models, and workflows will be linked to the Ocean Biogeographic Information System and Seabed 2030 to benefit a broad range of communities and stakeholders.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 06, 2024
- Source ID
- FA95502310494
Entities
People
- Xi Peng
Organizations
- Air Force Office of Scientific Research
- Office of the Secretary of Defense
- University of Delaware