We proposed a general method for domain adaptation which blends features from source and target datasets by a set of linear operations such a cross-correlation and convolution. We explore the viability of the approach by transferring knowledge from training on gray-scale MNIST dataset to colored MNIST-M dataset. ArXiv, 2021
Research project from my internship in CGG, Crawley, UK. Under supervision of Song Hou I explored benefits and pitfalls of using GANs for seismic data reconstruction. In the nutsell, high perceptual realism of reconstructed seismic data is not sufficient for real-world application. Physics also should be constrained. First EAGE Digitalization Conference and Exhibition, Nov 2020
In synthetic framework we explore capability of a simple neural network to reconstruct low-frequency components of land data. SEG Technical Program Expanded Abstracts 2020
We apply an iterative style transfer approach from image processing to produce realistically textured subsurface models based on synthetic prior models. These realistically textured models to be used in training datasets for machine learning applications in geophysics. Presented in Sep 2019, on SEG Annual meeting in San Antonio, USA
- 14/09/20: Started internship as Geoscience (ML) Intern at ExxonMobil, Houston, USA.
- 01/07/20: Abstract "Extrapolating low-frequency prestack land data with deep learning" got accepted for an oral presentation at SEG 2020.
- 01/03/20: Abstract "Deep Learning for Seismic Data Reconstruction: Opportunities and Challenges" got accepted for an oral presentation at EAGE Digital 2020.
- 18/09/19: Vladimir Kazei presented our work on style transfer for generation of realistic subsurface models at SEG Annual Meeting 2019 in San Antonio, USA. Code available!
- 06/08/19: Started internship as Machine Learnin Engineer Intern at CGG, Crawley, UK.
- 13/06/19: Awarded SEG ExxonMobil Upstream Research Company scolarship for the year 2019-2020!