Search papers, labs, and topics across Lattice.
3
0
5
Zero-shot composed video retrieval can jump by over 35% absolute R@1 without any parameter updates鈥攑rovided you route frozen foundation models as a dynamic compute ladder rather than a uniform scoring function.
Forgetting a class can be done with up to 99% accuracy while preserving the performance of other classes, revealing a new frontier in machine unlearning.
Geospatial agents can now reason more effectively about satellite imagery thanks to a new framework that aligns models with verified multi-step tool interactions and explicit reasoning traces.