Session: How Can AI/Stat Models be Interpreted Physically (Joint with the AMS Committee on Probability and Statistics) (99th American Meteorological Society Annual Meeting)

Wednesday, 9 January 2019: 10:30 AM-12:00 PM

North 124B (Phoenix Convention Center - West and North Buildings)

Host: 18th Conference on Artificial and Computational Intelligence and its Applications to the Environmental Sciences

Cochairs:

Philippe Tissot, Texas A&M University−Corpus Christi, Conrad Blucher Institute, Corpus Christi, TX; Amy McGovern, University of Oklahoma, School of Computer Science, Norman, OK and Elizabeth A. Satterfield, NRL, Marine Meteorology Division, Monterey, CA

The session will tackle the use of AI and Stat models for physical interpretability of nonlinear systems as well as the interpretation of the models themselves. Nonlinear environmental systems cannot simply use linear analysis methods to understand their behavior. Can relevant information be obtained from successful AI/ML model development, or are the underlying processes themselves sufficiently complex as to not be decipherable? If the latter, insights from successful nonlinear model development could be misleading. Submissions are sought in the areas of AI and STAT methods used to tackle classification, deep learning, regression, and probabilistic distribution problems.

Papers:

10:30 AM

TJ17.1

Making the Black Box More Transparent: Understanding the Physical Implications of Machine Learning (Core Science Keynote)

Amy McGovern, University of Oklahoma, Norman, OK; and D. J. Gagne II, R. A. Lagerquist, E. Jergensen, and K. L. Elmore

Recorded Presentation

11:00 AM

TJ17.2

Physics-Informed Generative Learning to Emulate Unresolved Physics in Climate Models

Jinlong Wu, Virginia Tech, Blacksburg, VA; and K. Kashinath, A. Albert, M. Prabhat, and H. Xiao

Recorded Presentation

11:15 AM