Experience of applying convolutional neural networks to inverse problems of seismic exploration

The paper is devoted to the study of the possibility of using modern convolutional neural networks to solve problems of reconstructing the position of geological inclusions and estimating the scalar parameters of the models used based on seismic exploration data. Synthetic seismograms calculated by explicit-implicit grid-characteristic schemes are used to form training and validation samples. The paper considers two network architectures for joint machine learning problems and compares the results of the calculated estimates with single forecast models. A significant increase in forecast quality is demonstrated.