Vertical density slice VORTEX Workshop
Versatile Open Representations and Token Exchange for Fluid Simulations
Towards interoperable, physics-aware data representations for HPC-AI
- Date
- Format
- In-person workshop
From simulation data to reusable AI models
Foundation models for scientific computing need data they can learn from across codes and applications. Fluid simulations produce large volumes of spatiotemporal data across different physical regimes, numerical methods, meshes, geometries, and scales. That diversity makes it difficult to reuse models across simulations and experiments.
VORTEX brings together French and European experts in computational fluid dynamics, high-performance computing, scientific data representations, and AI. It will explore common formats and physics-aware tokenization methods for training and deploying models on heterogeneous simulation and experimental data.
Workshop topics
-
Data interoperability and standardization
Common representations for physical variables, geometries, structured and unstructured meshes, adaptive mesh refinement, boundary conditions, and physical metadata.
-
Physics-aware tokenization
Strategies for encoding multiscale, spatiotemporal fluid data into reusable representations while preserving physical structures, conservation properties, and relevant dynamics.
-
Multiphysics and multiple flow regimes
Representations for compressible, incompressible, turbulent, multiphase, and magnetohydrodynamic flows, across different levels of physical modeling and scale closure.
-
Scalable HPC–AI workflows
Integration with HPC simulation codes, in situ data processing, distributed training, scientific data formats, and AI foundation models.
-
Reliability and reproducibility
Quantifying representation and tokenization errors, assessing physical consistency, and tracking data provenance across datasets and AI workflows.
A shared research roadmap
Invited technical talks and interactive discussions will help participants identify shared scientific requirements, evaluate existing approaches, and define open research questions.
The roadmap will outline priorities, potential collaborations, and demonstrator use cases, with particular attention to reusable software and scalable approaches for the NumPEx ecosystem.
Workshop programme