Visual Para-Thinker: Divide-and-Conquer Reasoning for Visual Comprehension
概要
arXiv:2602.13310v2 Announce Type: replace-cross Abstract: Existing LLM test-time scaling laws emphasize the emergence of self-reflective behaviors through extended reasoning length. Nevertheless, this vertical scaling strategy often encounters plateaus in exploration as the model becomes locked int…
