自监督世界模型表征研究：可读不等于可控——预注册检验流体物理结构
Readable Is Not Steerable: Pre-Registered Tests of Physical Structure in Learned Fluid Representations
AI 导读
一项在 2D 退化 Navier-Stokes 流上进行的预注册研究测试自监督世界模型是否围绕物理量组织其 latent。JEPA 未通过预注册标准，自编码器的最慢无监督 latent 方向上 enstrophy 显著（最佳 r 0.80-0.86 对比未训练 0.38），加入 VICReg 反而降低该显著性。
正文 · 原文
Abstract:Self-supervised world models are often said to organize their latents around physical quantities. We test this on 2D decaying Navier-Stokes flow in three pre-registered studies, each tested once on held-out data. In Study 1, a joint-embedding predictive architecture (JEPA) trained to predict the flow fails the pre-registered criteria for organizing its latent around fluid quantities better than an untrained network of the same architecture (not supported); a JEPA trained on non-physical linear dynamics and a static invariance model that never sees time do as well or better. In Study 2, plain autoencoders that render the field make enstrophy prominent in their slowest unsupervised latent directions (best r 0.80-0.86 vs 0.38 untrained; fresh seeds, sealed data). Adding VICReg lowers this prominence in every comparison, though not by the pre-registered margin for vorticity, while raising the autoencoders' linear decodability (partially supported). In Study 3, we build steerable directions ("dials") by routing: the decoder gets overall scale only through a small conditioning code, while its other pathway carries a degraded copy of spectral shape. Label-free rules find a loudness dial in all 8 trained models and a shape dial in 3 of 5 fresh seeds on sealed data, against a pre-registered bar of 4; neither control shows a shape dial (partially supported). Across the studies, what a latent contains and what can be controlled through it come apart. In exploration, a supervised direction reading spectral shape almost perfectly moved nothing when steered. During the build, the loudness dial had to be found by its effect on the output (closest to a pure rescaling), not by data variance; for shape, a variance rule and an effect-constrained rule each worked in 3 of 5 seeds. The program used about 70 logged hours on one consumer GPU; all pre-registrations, deviations, code and outputs are public.
| Comments: | 30 pages, 9 figures. Code, pre-registrations and checkpoints: this https URL |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI); Fluid Dynamics (physics.flu-dyn) |
| Cite as: | arXiv:2611.05577 [cs.LG] |
| (or arXiv:2611.05577v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2611.05577 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Manvendra Modgil [view email]
[v1]
Tue, 3 Nov 2026 20:08:52 UTC (874 KB)
来源：arXiv cs.AI · arxiv.org