Reproduce “World-as-Graph: Relational World Modeling Through Latent Space Graphs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yaqi Yang, Shuo Huang, Yujin Huang, Fucai Ke, Jiatong Han, Xin Zheng
WAG predicts future states in object-centric videos by connecting latent object representations into a time-varying similarity graph and updating each object's memory with information from its neighbors and past. On synthetic CLEVRER video question answering, its batch-shared variant reaches 91.37% average per-question accuracy, versus 89.40% for C-JEPA with VideoSAUR; this does not establish real-world causal understanding.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.