TurnWise improves streaming video RL with turn-level credit
Original title: Turns Matter: Advancing Streaming Video RL via Turn-Level Credit
TurnWise assigns turn-level advantages using silence-completed prefix scoring to reduce redundancy in streaming video policy learning
Streaming video understanding requires repeated decisions to respond or observe. Existing approaches use trajectory-level feedback in GRPO. TurnWise uses turn-level credit assignment with silence-completed prefix scoring and turn-mean policy optimization. Improves performance without learned critic or extra rollouts. Outperforms prior RL methods.
- separates learning signal from earlier rewards
- response and waiting decisions receive equal weight
- no additional rollouts or learned critic needed
- outperforms prior approaches for streaming video understanding
From the original article
1 min read at huggingface.co
Streaming video understanding requires models to repeatedly decide whether to respond or continue observing. Recently, Group Relative Policy Optimization (GRPO) has been applied to this setting, with existing approaches emphasizing reward design while retaining a single trajectory-level advantage across turns.
By Shuai Dong, Kanghui Tian, Linghan Chen · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.