V-LUMEN: Visual Lookup Memory for Embedding Scaling in Vision-Language Models
Adds lookup-based visual memory to vision-language models, scaling capacity without dense compute.
Explainer · on the wire
The explainer for this story is still on the wire.
The claim, the mechanism, and the figures land here once generated.
Source · arXiv