Search papers, labs, and topics across Lattice.
University of Toronto, Vector Institute
3
0
6
WorldTrace redefines memory management in video world models, achieving up to 19.5% better episodic recall without any retraining.
Self-correcting models can achieve unprecedented fidelity and plausibility in generative tasks by actively learning from their own alignment errors.
Forget scaling laws: a single looped transformer block, iterated explicitly, crushes billion-parameter feed-forward networks at multi-view 3D reconstruction.
