作者位次 #3/6ACL 2026Poster· Natural Language Processing
Interleaved Latent Visual Reasoning with Selective Perceptual Modeling
Shuai Dong · Siyuan Wang · Xingyu Liu · Chenglin Li · Haowen Hou · zhongyu wei
摘要
Interleaved reasoning paradigms enhance Multimodal Large Language Models (MLLMs) with visual feedback but are hindered by the prohibitive computational cost of re-encoding pixel-dense images. A promising alternative, latent visual reasoning, circumvents this bottleneck yet faces limitations: meth...