Generative Universal Verifier as Multimodal Meta-Reasoner
ICLROral2026
TL;DR
We introduce *Generative Universal Verifier*, a novel concept and plugin designed for next-generation multimodal reasoning in vision-language models and unified multimodal models, providing the fundamental capability of reflection and refinement on visual outcomes during the reasoning and generation process. This work makes three main contributi...
Opening excerpt from the authors’ abstract. source
Read the paper
Topics
vision-language language model multimodal reasoning