 |
Center for Nonlinear Studies |
Using a new approach to unsupervised abductive learning on historical records of seismic events, we recover long-range spatio-temporal structure in global seismicity. Our approach is knowledge-free; we do not inject any a priori knowledge of seismic stress propagation dynamics, nor do we assume any model of rheology, or constrain the analysis in any way with an a priori expectation of the physics in play. Our technique allows us to consider the data, and only the data; and distill hidden predictive patterns in the observed seismicity. Free from the need to instill a priori models, there are no parameters to tune, or physical constants to set. Our results indicate causal connections between seismic dynamics observed in California to that on the eastern edge of the Pacific plate, and additionally such hidden connections is shown to exist between events temporally separated by nearly a decade. Our technique allows for short-term earthquake prediction as well, and we show that the area under the receiver-operating characteristics curve is significantly greater than half, establishing non-trivial performance above random decisions. In addition, our technique is not limited to seismic prediction, and we point out a few high-impact applications and preliminary results relating to modeling mutational dynamics of retroviruses, and price fluctuations in financial markets.
Host: Marian Anghel