Social-JEPA: Emergent Geometric Isomorphism in Independently Trained World Models

Haoran Zhang, Youjin Wang, Yi Duan, Rong Fu, Dianyu Zhao, Sicheng Fan, Shuaishuai Cao, Wentao Guo, Xiao Zhou · 2026

Read Social-JEPA: Emergent Geometric Isomorphism in Independently Trained World Models by Haoran Zhang, Youjin Wang, Yi Duan, Rong Fu, Dianyu Zhao, Sicheng Fan, Shuaishuai Cao, Wentao Guo, Xiao Zhou (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2603-02263.txt.

Description

Haoran Zhang, Youjin Wang, Yi Duan, and colleagues study separately trained world models in this 2026 paper. World models squeeze rich sensory input into compact codes that predict what comes next. Two agents learned such models from different views of the same place. They shared no weights and never coordinated during training.

After training, the two inner spaces turned out to match in shape. A near-perfect linear map, preserving distances, links one space to the other. The match held despite large view shifts and little overlap in raw pixels. A classifier trained on one agent moved to the other with no retraining. Moving knowledge that way also sped later learning and cut total compute. Shared prediction goals seem to force shared geometry.

Catalog record

CREATORHaoran Zhang, Youjin Wang, Yi Duan, Rong Fu, Dianyu Zhao, Sicheng Fan, Shuaishuai Cao, Wentao Guo, Xiao Zhou
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE60774
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2603.02263

Browse the full library · About the Tower of Babel