Machine Learning Applied to Discern Fault Characteristics

Thursday, October 25, 2018 - 10:05am - 10:55am
Keller 3-180
Paul Johnson (Los Alamos National Laboratory)
We know that fundamental information is embedded in many conventional geophysical continuous data streams but that our ability to extract this information is limited by several factors: (a) it is often not apparent to human inspection and/or (b) when it is apparent, it requires a protracted expert analysis to interpret; (c) and/or we may be asking the wrong question of the data (human bias or ignorance). Our strategy is based on lessons learned from previous attempts to apply data-analytics, such as failed attempts at earthquake prediction. While machine learning offers a powerful path to extracting information rapidly from complex datasets, it must be strongly coupled to a fundamental understanding of the physical system to be meaningful and believable. I will describe an overview of our work applying continuous geophysical data streams. The primary focus of the work described will be on faulting and earthquakes, however we are turning our attention to many scales and many geological phenomena.