Causal JEPA at ICML
Author: @hazel_heejeong | Date: 2026-05-03
Announcement that Causal JEPA will be presented at ICML, with a recording from Stanford CS25 now available. Causal JEPA extends the JEPA framework with causal structure, bridging self-supervised representation learning and causal reasoning.
Key Claims
- Causal JEPA accepted at ICML 2026
- Talk recording available from Stanford CS25 seminar
- Extends JEPA (Joint Embedding Predictive Architecture) with causal mechanisms
- Bridges self-supervised learning and causal inference
Takeaways
- JEPA-family architectures gaining academic traction beyond Meta’s original work
- Causal structure in representation learning could improve generalization and reasoning
- Stanford CS25 continues to be a key venue for cutting-edge architecture talks