Search papers, labs, and topics across Lattice.
3
0
5
1
AdaCodec cuts time-to-first-token from 9.26 seconds to just 1.62 seconds while outperforming traditional per-frame RGB encoding methods.
VLMs are surprisingly brittle: swapping text for a semantically identical image tanks performance, but a new data curation method, LoMo, can fix it.
Random projections in continual learning don't have to be random: carefully guiding them with target-aligned data beats the SOTA.
