Search papers, labs, and topics across Lattice.
4
8
8
7
Teacher Prediction Refinement Distillation (TPRD), a plug-and-play module that refines teacher predictions before distillation by exploiting stage-wise prediction information, and Maximum Dark Knowledge Preservation (MDKP), which selectively refines target-class logits while retaining non-target relations.
Current video generation models struggle with visual reasoning, with the best achieving only 51% accuracy on a new benchmark designed to probe their capabilities.
Reconstructing realistic 3D hand avatars from messy, real-world video just got a whole lot better thanks to a new method that explicitly models and suppresses visual "noise" like motion blur and object interactions.
A 7B parameter agent can now rival the performance of models 4-10x larger at autonomous computer operation, thanks to innovations in data generation, reinforcement learning, and model enhancement.
