In the course of analyzing complex geophysical systems, the apeture and density of data acquisition is limited, and we are inevitably faced with the challenge of inverting subsurface properties and imaging subsurface structures under partial information. In this small data regime, the vast majority of state-of-the-art machine learning techniques (e.g., deep/convolutional/recurrent neural networks) are lacking robustness and fail to provide any guarantees of convergence. In this project, I apply physics informed neural networks (PINNs) that are trained to solve supervised learning tasks while respecting any given law of physics described by general nonlinear partial differential equations (PDEs) to seismic modeling, inversion and imaging.
Physics informed neural networks
The recently proposed physics informed neural networks (PINNs), which are trained to solve supervised machine learning tasks while respecting the given law of physics, sheds a light on full wave simulation that is free of numerical instability and artifical boundary.
