TurnWise改进流式视频强化学习
AI 总结
TurnWise是一种面向流式视频交互的GRPO框架,将强化学习信用分配到每次“回复或继续观察”的决策,而非让整条轨迹共享同一个优势值。作者称,其静默完成前缀评分和轮次均值策略优化无需学习型评论家或额外采样,即可改善模型行为,并在实验中超过此前的强化学习方法。
深度见解
TurnWise为流式视频模型的逐轮决策训练提供了具体方案,但目前的性能提升结论仅来自作者实验。
Computer Science > Computer Vision and Pattern Recognition
arXiv:2611.06728 (cs)
[Submitted on 4 Nov 2026]
Title:Turns Matter: Advancing Streaming Video RL via Turn-Level Credit
Authors:Shuai Dong, Kanghui Tian, Linghan Chen, Wenqi Shao, Hongrui Wu, Haowen Hou, Yongfu Zhu, Xuehui Wang, Ziyue Wang, Wenpu Liu, Weichu Xie, Yuqi Xu, Congcong Wang, Kaiwen Tuo, Caoyuan Ma, Taiqiang Wu, Ji Zhao, Wenzheng Chang, Dianyi Wang, Siyuan Wang, Zhongyu Wei, Yu Qiao, Jiaqi Wang
View a PDF of the paper titled Turns Matter: Advancing Streaming Video RL via Turn-Level Credit, by Shuai Dong and 22 other authors