Track the frontier, from condensed matter to cosmology.
Search papers, labs, and topics across Lattice.
5
0
7
2
Current omnimodal video embeddings struggle with physical reasoning, yet specific physics can be recovered and leveraged to significantly improve the physical fidelity of generated videos.
Pixel Linguist II not only sets new benchmarks in multilingual visual text understanding but also maintains performance under 80% visual token compression.
AtomiMed achieves a strikingly higher correlation with human radiologist evaluations, transforming how we assess clinical report accuracy.
A tailored vision-language framework boosts diagnostic accuracy on 3D CT imaging by over 5% while enhancing zero-shot reliability through innovative semantic alignment techniques.
LLMs can learn to tailor their queries for different retrievers, revealing that one-size-fits-all strategies are ineffective in retrieval-augmented generation.