Search papers, labs, and topics across Lattice.
3
0
6
10
AMVL eliminates the critical train-inference mismatch in MLLMs, leading to substantial improvements in reasoning performance across multimodal tasks.
LoRA's catastrophic forgetting problem can be solved by explicitly decoupling its factor updates and pushing new feature representations away from old ones, leading to state-of-the-art continual learning performance.
VLMs can now access the full visual hierarchy on-demand, thanks to a new cross-layer injection method that dynamically bridges vision encoders and LLMs.
