Search papers, labs, and topics across Lattice.
University of Maryland
3
0
6
Treating natural state trajectories as self-supervised "actions" turns static scientific emulators into interactive, counterfactual world models without requiring manually labeled intervention data.
Generative modeling on Lie group manifolds reveals that traditional Euclidean approaches can obscure critical dynamics, leading to significant errors in data representation.
Evaluations reveal that current models struggle with long-range narrative integration and cultural reasoning, highlighting a critical gap in video understanding capabilities.
