Determination of earthquake focal mechanism via multi-task learning.

A multi-task learning-based focal mechanism network (MTFMN) is proposed for calculating parameters of the focal mechanism of earthquakes by regression with incorporating expert prior knowledge. The model automatically learns feature representations of seismic waveforms and transforms the inversion task of the focal mechanism into multi-task learning. Experimental results suggest that MTFMN outperforms traditional methods in the task of earthquake focal mechanism and improves the accuracy of parameter estimation of focal mechanism. In addition, comparative experiments demonstrate MTFMN's enhanced robustness and generalization capabilities compared to other methods. Our proposed methodology presents a more precise regression approach for focal mechanism inversion, with the potential to provide a better understanding of seismic events. • Solving the focal mechanism using an end-to-end model. • Incorporating expert knowledge in deep learning models. • A soft-shared multi-task learning model is used to improve model performance. [ABSTRACT FROM AUTHOR]