Search papers, labs, and topics across Lattice.
1
0
2
3
Streaming video MLLMs perform far better when historical context is actively internalized into evolving latent tokens rather than queried as passive external visual buffers.
