Research map / Multi-agent systems and games / Cooperative multi-agent RL
VideoChat-M1: Collaborative Policy Planning for Video Understanding via Multi-Agent Reinforcement Learning
Boyu Chen, Zikang Wang, Zhengrong Yue, Kainan Yan, Chenyun Yu, Yi Huang, Zijun Liu, Yafei Wen, Xiaoxin Chen, Yang Liu, Peng Li, Yali Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add collaborative multi-agent policy planning with 4 agents instead of 1 (Qwen3-8B) (ablation) | VideoChat-M1 · LongVideoBench | accuracy: 61.9 -> 82.3 | Improved |
| add result reward R_res (ablation) | VideoChat-M1 MARL · Video-Holmes | accuracy: 32.4 -> 60.5 | Improved |
| add collaboration (process) reward R_col (ablation) | VideoChat-M1 MARL · LongVideoBench | accuracy: 81.1 -> 82.3 | Improved |
| add format reward R_format (ablation) | VideoChat-M1 MARL · LongVideoBench | accuracy: 82.0 -> 82.3 | Improved |
| add agent dropout (random DAG communication topology) (ablation) | VideoChat-M1 MARL · LongVideoBench | accuracy: 79.9 -> 82.3 | Improved |
| add MARL after SFT (ablation) | VideoChat-M1 · LongVideoBench | accuracy: 75.9 -> 82.3 | Improved |
| add SFT warm-up before MARL (ablation) | VideoChat-M1 · LongVideoBench | accuracy: 80.2 -> 82.3 | Improved |
| use heterogeneous agent architectures instead of 4 identical Qwen3-8B agents (ablation) | 4-agent CPP group · LongVideoBench | accuracy: 79.2 -> 82.3 | Improved |
| MARL-trained small agent group instead of untrained GPT-4o/DeepSeek-R1 groups under CPP (ablation) | CPP agent group · LongVideoBench | accuracy: 75.9 -> 82.3 | Improved |
| full-parameter finetuning instead of LoRA (ablation) | VideoChat-M1 · LongVideoBench | accuracy: 81.2 -> 82.3 | No clear effect |
| majority vote instead of a single designated agent deciding (ablation) | VideoChat-M1 · LongVideoBench | accuracy: 81.6 -> 82.3 | Improved |
| increase homogeneous agents beyond four (ablation) | CPP agent group · Qwen3-8B homogeneous groups | accuracy | No clear effect |
Closest papers by meaning
- ReAgent-V: A Reward-Driven Multi-Agent Framework for Video Understanding (NeurIPS 2025 · similarity 0.90)
- Find, Fix, Reason: Context Repair for Video Reasoning (ICML 2026 · similarity 0.89)
- EVA: Efficient Reinforcement Learning for End-to-End Video Agent (CVPR 2026 · similarity 0.89)
- ReWatch-R1: Boosting Complex Video Reasoning in Large Vision-Language Models through Agentic Data Synthesis (ICLR 2026 · similarity 0.88)
- VideoWebArena: Evaluating Long Context Multimodal Agents with Video Understanding Web Tasks (ICLR 2025 · similarity 0.87)
- Video-R1: Reinforcing Video Reasoning in MLLMs (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
