AI Paper Atlas

Research map / Multi-agent systems and games / Cooperative multi-agent RL

DECOR: Learning to Decompose and Collaborate in Deep Search via Multi-Agent Reinforcement Learning

Ruiqing Chen, Zekun Zhang, Gong-Duo Zhang, Lihong Gu, Lin Zhou

ICML 2026 · Multi-agent systems and games › Cooperative multi-agent RL · approach: MARL benchmarks and grouping

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 changedWhereMeasuredEffect
use hybrid reward (team F1 + judge + role-specific) instead of F1-only, EM-only or LLM-Judge-only (ablation)DECOR · 2WikiMQALLM-Judge accuracyImproved
vary reward weight alpha between team and role rewards (ablation)DECOR · 2WikiMQALLM-Judge accuracy: peak at alpha = 0.6Mixed
increase retrieved documents top-k from 5 to 10 (ablation)Search-R1 · HotpotQALLM-Judge accuracy: 59.1 -> 56.4Hurt
increase retrieved documents top-k from 5 to 10 (ablation)DECOR · HotpotQALLM-Judge accuracy: 59.8 -> 60.1Improved
increase retrieved documents top-k from 5 to 10 (ablation)DECOR · 2WikiMQALLM-Judge accuracy: 62.0 -> 62.2No clear effect
add Qwen3-Reranker (ablation)DECOR · 2WikiMQALLM-Judge accuracy: 62.0 -> 65.8Improved
add Qwen3-Reranker (ablation)DECOR · HotpotQALLM-Judge accuracy: 59.8 -> 60.5Improved
replace static Wikipedia corpus with live Google Search API (ablation)DECOR · 2WikiMQALLM-Judge accuracy: 62.0 -> 68.4Improved
replace static corpus with live Google Search API (ablation)DECOR · HotpotQALLM-Judge accuracy: 59.8 -> 62.6Improved
use multi-agent DECOR instead of monolithic Search-R1deep search agents · 2WikiMQA, BamboogleF1: 2WikiMQA 61.8 (R1-Searcher) vs 61.2; Bamboogle 52.5 (Search-R1) vs 50.3Mixed
use DECOR on MuSiQuedeep search agents · MuSiQueF1: 24.9 (SimpleDeepSearcher) -> 23.8No clear effect

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy