page content

W-05 Machine Learning vs Conventional Approaches/Workflows in Applied Geophysics – Challenges, Values, and Where We Are Heading?

Through the support of the Research Committee

Thursday, 30 September 2021, 8:30 a.m.–5:00 p.m.  |  Denver, Colorado

short course image

Course Content

There is a significant amount of applications of machine-learning algorithms published in the area of applied geophysics in the past few years. While these pioneering works show great potential of the algorithms, many of them are straightforward applications of machine-learning methods such as neural networks for various geoscience functionalities and workflows. Some of these methods, especially those for seismic processing applications, do not outperform the conventional methods. We certainly do not want apply machine learning for the sake of just applying machine-learning algorithms. We would like to show the value that machine learning brings for our works. In this workshop, we will discuss machine-learning applications for applied-geophysics problems. However, we will put an emphasis on the value of machine learning over conventional approaches and workflows. Discussion about rigorous comparisons and benchmarks between the existing conventional methods and the machine-learning-based methods will be the focus of the workshop.

More specifically, we will discuss the value propositions of machine-learning-based methods/workflows such as:

  • Efficiency improvement in terms of complexity and scalability
  • Accuracy, robustness, and generalization improvements
  • Long-existing problems that cannot be solved using conventional approaches/workflows

Session 1
  • Artificially intelligent waveform inversion using neural network functions: Tariq Alkhalifahm KAUST
  • Deep-learning-based Artificial Bandwidth Extension: Learning low-frequency reconstruction from seabed seismic to enhance towed-streamer FWI: Mehdi Aharchaou, ExxonMobil
  • Panel Discussion
Session 2
  • Seismic image conditioning using deep learning: Hiren Maniar, Schlumberger
  • Image-to-image processing and well tie with deep neural networks: Norman Ettrich, Fraunhofer ITWM
  • Producing high-resolution seismic data from well logs with deep learning methods: Nam Pham, Aramco
  • Panel Discussion
Session 3
  • A 'labels-first' approach to deep learning for seismic interpretation: Matt Morris, Bluware
  • Machine augmented seismic interpretation: Diego Hernandez: ExxonMobil
  • Panel Discussion
Session 4
  • Physics-driven machine learning-enabled wellsite automatic borehole sonic shear processing: Lin Liang, Schlumberger
  • Neural architecture search for inversion: Cheng Zhan, Microsoft
  • DAS-based high resolution VSP with deep learning: Vladimir Kazei, Aramco
  • Panel Discussion


IMAGE '21 Workshop Pass
Non-Member Fee: $229
Member Fee: $129
Non-Member Students: $59
Member Students: $39

All Postconvention Workshops will be virtual. The Workshop Pass includes access to any or all postconvention workshops.


W-05 Machine Learning vs Conventional Approaches/Workflows in Applied Geophysics – Challenges, Values, and Where We Are Heading?
Colorado Convention Center
700 14th St
Denver, Colorado 80202
United States


Wenyi Hu Wenyi Hu 2021-2021 Co-Organizer, AGT
Weichang Li Weichang Li 2021-2021 Co-Organizer, Aramco
Lin Liang Lin Liang 2021-2021 Co-Organizer, Schlumberger
Mehdi Aharchaou Mehdi Aharchaou 2021-2021 Co-Organizer, XOM
Aria Abubakar Aria Abubakar 2021-2021 Co-Organizer, Schlumberger