Combining Gradient Optimization and Machine Learning Methods for Inverse Problems in Layered Heterogeneous Media

The paper considers the combination of classical gradient optimization methods and convolutional neural networks for solving the inverse problem of obtaining the elastic properties of layered heterogeneous media. The convolutional neural network is trained using synthetically generated numerical data. The network is used to provide an initial approximation for the further work of the gradient optimization method. The paper considers different approaches to neural network training and compares the performance of gradient optimization when starting from their predictions.