Calculating the gradient is an essential step for the iterative deterministic inversion method, where the convergence virtually depends on the accuracy and efficiency of the gradient estimation. However, in conventional methods, the estimation of the gradient may incur a high computational cost and the solution is easy to be stuck into a local minimum. In this paper, we propose a Deep Learning based method to accelerate the gradient calculation and leverage prior information introduced by the training set. We implement an optimization framework by using the Recurrent Neural Network (RNN) and train the network based on a global criterion. The feasibility of the proposed approach is verified by solving a magnetotelluric (MT) inverse problem.
Presentation Date: Wednesday, October 14, 2020
Session Start Time: 9:20 AM
Presentation Time: 10:35 AM
Location: Poster Station 1
Presentation Type: Poster