HP-JEPA: Hierarchical Partitioning for Multi-Resolution Graph Joint-Embedding Predictive Learning

Ruichen Xu, Jingxiang Qu, Wenhan Gao, Jiaxing Zhang, Linsey Pang, Ravid Shwartz Ziv, Yann LeCun, Yuefan Deng
August 2026 arXiv, 2026
PDF arXiv

Abstract

HP-JEPA learns graph representations by predicting masked targets in latent space across a hierarchy of graph partitions. Combining features from coarse and fine partitions lets downstream models use local, regional, and global structure. The approach is evaluated on graph classification and regression tasks.

Type
Preprint
Publication
arXiv preprint
Representation Learning Graph Learning JEPA Self-Supervised Learning
Ravid Shwartz Ziv
Ravid Shwartz Ziv
AI Researcher

AI researcher at Meta MSL with a background in information theory and computational neuroscience, working on world models, memory, and compression. Former Assistant Professor and Faculty Fellow at NYU. My industry experience spans Wand AI, Intel, Google AI, and Wikipedia.

Related