PI-JEPA: Label-Free Surrogate Pretraining for Coupled Multiphysics Simulation via Operator-Split Latent Prediction
arXiv:2604.01349v4 Announce Type: replace Abstract: Reservoir simulation workflows face a fundamental data asymmetry: input parameter fields (geostatistical permeability realizations, porosity distributions) are free to generate in arbitrary quantities, yet existing neural operator surrogates require large corpora of expensive labeled simulation trajectories and cannot exploit this unlabeled structure. We introduce \textbf{PI-JEPA} (Physics-Informed Joint Embedding Predictive Architecture), a surrogate pretraining framework that trains \emph{without any completed PDE solves}, using masked late
The increasing complexity of multiphysics simulations and the growing availability of unlabeled data are driving the need for more efficient and less data-intensive AI models.
This development significantly lowers the barrier to entry for AI-driven simulation, enabling faster and cheaper exploration of complex systems without the need for extensive labeled datasets.
Traditional neural operator surrogates required large, expensive labeled simulation trajectories, but PI-JEPA changes this by allowing pretraining with readily available unlabeled input parameter fields.
- · AI model developers
- · Engineering and scientific research
- · Industries relying on complex simulations (e.g., energy, manufacturing, climate
- · Cloud computing providers
- · Traditional high-cost simulation services
- · Specialized data labeling services for simulations
Accelerated discovery and optimization processes across various scientific and engineering disciplines due to more accessible and efficient simulation.
Increased demand for computational resources and specialized AI hardware to run these advanced pretraining and simulation frameworks.
Potential for new AI-driven design paradigms that integrate simulation much earlier and more continuously into product development lifecycles.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG