Metraipolis › archive › dc_49a761124418
Frozen in Time: How a Physics Foundation Model Forgets Under Sim-to-Real Adaptation — arXiv:2610.40857
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.40857
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-03T08:11:38.727Z
- SHA-256 of the archived text
277294d99cf314f17b7f6a25b77b30f28802fc0f5340d29a1a4fb9539b1a8c1a- Stable URL of this page
https://metraipolis.com/documento/dc_49a761124418
How to cite
Frozen in Time: How a Physics Foundation Model Forgets Under Sim-to-Real Adaptation — arXiv:2610.40857 — arXiv (en), https://arxiv.org/abs/2610.40857 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_49a761124418 (Metraipolis archive dc_49a761124418, sha256 277294d99cf314f17b7f6a25b77b30f28802fc0f5340d29a1a4fb9539b1a8c1a)
BibTeX
@misc{dc49a761124418,
title = {Frozen in Time: How a Physics Foundation Model Forgets Under Sim-to-Real Adaptation — arXiv:2610.40857},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.40857},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_49a761124418; sha256 277294d99cf314f17b7f6a25b77b30f28802fc0f5340d29a1a4fb9539b1a8c1a},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.40857. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Scientific foundation models (SciFMs) have achieved remarkable performance on a wide range of PDEs, including the ability to efficiently adapt to unseen data distributions with additional finetuning. As with all neural architectures, finetuning such models comes with the risk of catastrophic forgetting, whereby adapting to new tasks degrades prior knowledge of the model. We study catastrophic forgetting in Walrus-1.3B, a SciFM for continuum dynamics, over the course of a two-stage sim-to-real adaptation: first finetuning on simulated flow around a cylinder and then adapting to experimental data of the same system. We find that finetuning on the simulations rapidly degrades rollout performance on the original pretraining datasets, with one-step errors increasing up to 35x after just 20 epochs. Subsequent adaptation to real data further erodes performance on the simulated task. We characterize the potential drivers of forgetting, such as dataset dependence and task representation similarity. We then show that replaying as little as 1% simulated data during real-data finetuning recovers most of the lost simulation performance while retaining effective adaptation to the experimental dynamics. Our results identify catastrophic forgetting as a practical obstacle to sim-to-real adaptation of SciFMs and show that lightweight replay strategies significantly improve retention. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG) ; Fluid Dynamics (physics.flu-dyn). Pagina del lavoro: https://arxiv.org/abs/2610.40857 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-11-03T08:02:38.893Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).