Preprint
Aug 2026
Messages, Not Tokens: Grounded Coresets for Faithful VLM Compression
Fidelity visual compression is formulated as constructing a compact coreset for decoder messages, and Grounded Message Coreset Pruning is introduced which jointly allocates support across query-grounded, appearance, and coordinate-aware evidence, then transports discarded states into selected representatives at their original multimodal positions before physical compaction and native attention resume.
Long Qian, Jiaqi Wei, Bingke Zhu et al.
· 0 citations