Lightning Talk: Seismic Data to Subsurface Models with OpenFWI
summary
Neural network directly maps seismic shot gathers to velocity models using the OpenFWI InversionNet UNet-style architecture trained on roughly 150,000 shot gathers and 30,000 subsurface models (120k/24k train, 25k/5k val, 5k/1k test). Training fine-tunes on a CPU cluster with 4th Gen Xeon with AMX using one node and 32 cores, converging over 50 epochs in ~4 hours and producing close matches to ground truth without any physics-based starting model.
AI-written summary · about
chapters (8)
- 0:00Introduction to seismic FWI
- 0:20Importance of seismic imaging
- 1:56Challenges in traditional FWI
- 3:20Field data and methodology
- 4:45Training data and architecture
- 5:52Results and CPU performance
- 7:00Conclusions and resources
- 8:56Q&A and closing remarks
related talks
45mSolving Inverse Problems with Deep Learning
25m2D CNN-based Nearshore Bathymetry Inversion from Synthetic Imagery
17mAI Weather Forecasting
25mConvergence of AI, Simulations and HPC
42mH2O Deep Water with Python early sneek
31mScientific Machine Learning With NVIDIA PhysicsNeMo
27mFoundation Models: Blueprints for success
30mRay Across the Stack Fine Tuning Qwen and Serving 3D Generation at Autodesk with AWS Trainium and In