AI
PDE-JEPA: Predictive Representation Learning of Latent Dynamics Modeling for Parametric PDEs
Publication: Hugging Face Daily Papers
Author: Not available in collected metadata
Source published: 27/09/2026, 20:00:00 UTC
Briefing created: 05/10/2026, 03:26:13 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 05/10/2026, 03:18:24 UTC.
Briefings
This article presents PDE-JEPA, a method for parametric PDEs that uses predictive representation learning to capture underlying dynamics rather than just reconstructing observed fields. The framework trains an encoder via masked-latent prediction and employs a geometry projector to align latent trajectory geometry with physical field evolution. Experiments on nine PDE benchmarks show the method outperforms SOTA by 33.4% in-distribution and 51.4% when extrapolating to unseen governing parameters.
Key points
- PDE-JEPA introduces a predictive representation learning approach for parametric PDEs, utilizing a geometry projector to align latent trajectory geometry with physical field evolution.
- The framework achieves a 33.4% average improvement in-distribution and a 51.4% average improvement when extrapolating to unseen governing parameters on nine PDE benchmarks.
Context and limitations
Results are based on nine PDE benchmarks; specific performance metrics for unseen parameters are limited to the reported 51.4% extrapolation improvement.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 05/10/2026, 03:26:13 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeierv7cy7aqkwof32blqkwllopbvx2eqpuyvzt7wad7ybw4ym3jdzi