Search papers, labs, and topics across Lattice.
4
0
5
10
Multimodal context learning remains critically underdeveloped, with top models barely scratching the surface of effective performance.
OrthoPilot outperformed seasoned orthopaedic experts in diagnostic reasoning, achieving a 10.6% increase in management success for complex musculoskeletal cases.
BETA achieves state-of-the-art continual learning performance with just 0.05 million trainable parameters, outperforming traditional methods by 180–3000 times in parameter efficiency.
Ditch the slow, iterative zooming during MLLM inference: Region-to-Image Distillation lets you bake those agentic zooming benefits directly into a single forward pass.
