TurnWise: Turn-Level Credit Assignment for Streaming Video RL

A paper from Fudan, JD and others proposes TurnWise, a turn-level GRPO framework that assigns credit to each decision of when to respond versus keep watching in streaming video understanding. This fixes vanilla GRPO's flaw of sharing the same positive signal across all turns.

2026-11-05 ~ 2026-11-06 · 2 related posts

• TurnWise: Turn-Level Credit Assignment Advances Streaming Video RL — Shuai Dong · 2026-11-05
• TurnWise: turn-level GRPO improves when-to-answer in streaming video RL — _akhaliq · 2026-11-06