Social-JEPA: Emergent Geometric Isomorphism
cs.CV, cs.AI
Submitted: 2026-02-28
Updated: 2026-08-29
Comments: arXiv admin note: This submission has been withdrawn by arXiv administrators due to incorrect authorship. Author list truncated
License: http://creativecommons.org/licenses/by/4.0/
The gist: World models compress rich sensory streams into compact latent codes that anticipate future observations.
Terminology
Abstract
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.
Sources
- Understanding intermediate layers using linear classifier probes
- V-JEPA 2: Self-Supervised Video Models Enable Understanding, Prediction and Planning
- VICReg: Variance-Invariance-Covariance Regularization for Self-Supervised Learning
- Revisiting Feature Prediction for Learning Visual Representations from Video
- Distilling the Knowledge in a Neural Network
- The Kinetic Fokker-Planck equation in a domain: Ultracontractivity, hypocoercivity and long-time asymptotic behavior
- World Models
- The information bottleneck method
Related papers
- Loss Knows Best: Detecting Annotation Errors in Videos via Loss Trajectories
- AnchorWeave: World-Consistent Video Generation with Retrieved Local Spatial Memories
- Benchmarking the Robustness of Foundation Models for Mammography under Domain Shift
- MambaX-Net: Dual-Input Mamba-Enhanced Cross-Attention Network for Longitudinal MRI Segmentation
- TeleOCR: Navigating Document Parsing Across Digital and Camera-Captured Documents
- A Survey on Efficient Vision-Language-Action Models