Social-JEPA: Emergent Geometric Isomorphism
Demonstrates that predictive learning imposes strong geometric regularities, offering a lightweight path to interoperability for decentralized vision systems.
Two agents trained with predictive world models from different viewpoints develop latent spaces related by an approximate linear isometry, enabling zero-shot classifier transfer and reduced compute via distillation-like migration.
World models compress rich sensory streams into compact latent codes that anticipate future observations. We let separate agents acquire such models from distinct viewpoints of the same environment without any parameter sharing or coordination. After training, their internal representations exhibit a striking emergent property: the two latent spaces are related by an approximate linear isometry, enabling transparent translation between them. This geometric consensus survives large viewpoint shifts and scant overlap in raw pixels. Leveraging the learned alignment, a classifier trained on one agent can be ported to the other with no additional gradient steps, while distillation-like migration accelerates later learning and markedly reduces total compute. The findings reveal that predictive learning objectives impose strong regularities on representation geometry, suggesting a lightweight path to interoperability among decentralized vision systems. The code is available at https://anonymous.4open.science/r/Social-JEPA-5C57.