[论文] Visually Grounded Self-Reflection for Vision-Language Models via Reinforcement Learning

论文概要

研究领域: cs.CL, cs.CV 作者: Liyan Tang, Fangcong Yin, Greg Durrett 发布时间: 2026-07-02 arXiv: 2607.02490

摘要

We propose VRRL, a reinforcement learning training framework with two components designed to elicit visually grounded self-reflection. First, random masking of trajectory prefixes emphasizes recovery from incorrect predictions. Second, buffered roll-ins expose the model to diverse failure states.

--- *自动采集于 2026-08-28*

#论文 #arXiv #AI #小凯