Details of the Abstract
| Title of paper | Generalized Data-Driven Radio Magnetotelluric Inversion using Convolutional Neural Network |
| List of authors | Ghosal, K., Singh, A., Gupta, D. |
| Affiliation(s) |
Department of Applied Geophysics, Indian Institute of Technology (Indian School of Mines), Dhanbad, India Department of Applied Geophysics, Indian Institute of Technology (Indian School of Mines), Dhanbad, India Transmute AI Lab, Indian Institute of Technology (Indian School of Mines), Dhanbad, India |
| Summary |
Machine learning techniques are considered a suitable option for geophysical data inversion. However, ma- chine learning algorithms are very useful for inversion of geophysical data. For a data-driven approach, these algorithms work on the assumption that the statistical properties of the training and testing dataset follow an in- dependent and identical distribution (IID). However, such assumptions are not met in the geophysical field data and the solutions obtained using these algorithms may provide erroneous solutions. We propose a strategy for the inversion of RMT data where the training data was generated for resistivity models based on the Gaussian Random Fields (GRFs). Once the models were created, forward responses in terms of resistivity and phase were generated using a finite-difference based algorithm. A Convolutional neural network was trained in a supervised manner. The trained network was tested for various Out of Distribution (OOD) models including the checkerboard model. The proposed approach overcomes the need to retrain the neural network for different OOD samples. Hence introducing generalization ability in a data-driven supervised learning framework. |
| Session Keyword | 2.0 EM theory, modelling and Inversion |
| File upload |
2.0_generalized_data-driven_r_ghosal_02.pdf
|