For its second webinar of 2025 dedicated to Machine Learning, the EAGE Local Chapter Paris proposes you to deepen the topic with examples of Machine learning Geophysical applications selected from recent industry projects and state of the art university research work.
Speakers: Ammar Ahmad (SLB), Yuke Xie (Mines Paris-PSL), Nigar Alishzada (UFAZ)
Agenda
Performing Intelligent and Agile Field Development Planning through Geophysical & Geoscience Machine Learning Applications
By Ammar Ahmad, Subsurface Domain Lead UK & Europe at SLB
In this talk we will focus on several geophysical and geoscience machine learning applications which not only fast track the field development projects but also bring efficiency through intelligence of ML and AI workflows. The value proposition for deploying machine learning applications to field development planning not only brings prediction improvements but also contributes towards integrated workflows which transition these projects into dynamic evergreen models. The machine learning applications which widely range across geophysical applications with integration into subsurface reservoir modeling help to develop data conditioned models which have provided far greater improvements on predictability and improvements compared to conventional approaches. The ability to fast track such projects with these applications bring maximum efficiency from planning to delivery.
Generative prior for seismic imaging and uncertainty quantification
By Yuke Xie, PhD candidate in Geophysics at Mines Paris-PSL
To obtain high-resolution images of subsurface structures from seismic data, seismic imaging techniques such as Full Waveform Inversion (FWI) are essential tools. However, FWI involves solving a nonlinear and often non-unique inverse problem, leading to challenges such as local minima trapping and inadequate handling of inherent uncertainties. To address these challenges, we propose using deep generative models as the prior distribution of geophysical velocity parameters for stochastic Bayesian inversion. We compare four posterior sampling methods under generative prior. This framework offers insights into enhancing imaging quality while accounting for inherent uncertainties using machine learning.
Pre-Trained Models in Geoscience: Unlocking Insights with Transfer Learning
By Nigar Alishzada, computer Science teacher at the French-Azerbaijani University (UFAZ)
Rapid advancements in machine learning (ML) have opened new frontiers for tackling complex challenges in geoscience, from seismic interpretation to remote sensing and environmental monitoring. In this talk, we will explore the transformative potential of transfer learning—a technique that leverages pre-trained models from one domain and adapts them to solve problems in another. Transfer learning enables geoscientists to overcome data scarcity, reduce computational costs, and accelerate innovation by reusing knowledge from domains such as computer vision and natural language processing.
This talk aims to inspire collaboration between machine learning experts and geoscientists by showcasing how transfer learning can bridge the gap between data-rich and data-scarce domains.
