“It’s not like machine learning will solve all the problems. It’s not a magical tool.”
David Lubo-Robles highlights his award-winning paper that utilized novel machine learning methods to enhance interpretability in seismic volume data from the Gulf of Mexico.
Discover the power of two open-source tools – SHAP (Shapley Additive Explanations) and LIME (Local Interpretable Model-agnostic Explanations) – in enhancing the interpretability of machine models. David takes us through his team’s research that garnered an Honorable Mention for Best Paper in Interpretation. He also shares his journey into geophysics, driven by a fascination with the Earth and energy discovery.
Listeners will gain insight into the critical role of input quality in machine learning outcomes, the importance of balancing datasets, and the necessity of geoscientific validation. The episode also addresses common misconceptions about machine learning in geophysics, emphasizing the need for critical thinking and geological knowledge to apply these advanced techniques.
- 2:04 – How David discovered geophysics
- 4:32 – How SHAP and LIME improve machine learning for geophysics
- 6:00 – What to do when algorithms misclassify areas of interest
- 10:47 – A misconception common for machine learning in geophysics
- 13:37 – Sensory interpretation can be very subjective, even in the same area
- 15:00 – Managing uncertainty in the subsurface
- Read the award-winning paper: David Lubo-Robles, Deepak Devegowda, Vikram Jayaram, Heather Bedle, Kurt J. Marfurt, and Matthew J. Pranter, (2022), “Quantifying the sensitivity of seismic facies classification to seismic attribute selection: An explainable machine-learning study,” Interpretation 10: SE41-SE69.
- Local interpretable model-agnostic explanations (LIME) tool
- Shapley additive explanations (SHAP) tool
- Explore AASPI
- Interview transcript
Dr. David Lubo-Robles is a Postdoctoral Research Associate at the University of Oklahoma. David is a geophysicist interested in developing and applying innovative tools using machine learning, quantitative seismic interpretation, and seismic attribute analysis for oil and gas, geothermal reservoir characterization, hydrogen storage, and carbon capture, utilization, and storage (CCUS). His paper, “Quantifying the sensitivity of seismic facies classification to seismic attribute selection: An explainable machine-learning study,” was awarded Honorable Mention, Best Paper in Interpretation in 2022. David received his MS and Ph.D. in Geophysics at the University of Oklahoma.
Seismic Soundoff showcases conversations with geoscientists addressing the challenges of energy, water, and climate.
SEG creates these episodes to celebrate and inspire the geophysicists of today and tomorrow.
If you have episode ideas or feedback for the show or want to sponsor a future episode, email the show at [email protected].
This episode was hosted, edited, and produced by Andrew Geary at TreasureMint. The SEG podcast team is composed of Jennifer Cobb, Kathy Gamble, and Ally McGinnis.
Transcription and episode summary support provided by Headliner.