AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

MIRAGE: Scaling Test-Time Inference with Parallel Graph-Retrieval-Augmented Reasoning Chains

Kaiwen Wei, Rui Shan, Dongsheng Zou, Jianzhong Yang, Bi Zhao, Junnan Zhu, Jiang Zhong

AAAI 2026 · Large language models › Retrieval and knowledge · approach: Retrieval-augmented reasoning · 1 citations (OpenAlex)

Read the paper

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
graph-based multi-chain retrieval-augmented reasoning (MIRAGE) vs Search-o1QwQ-32B · ExplainCPEaccuracy (%): 80.7% -> 84.8%Improved
MIRAGE vs Search-o1QwQ-32B · GenMedGPT-5kGPT-4o rank: 3.3 -> 1.8Improved
MIRAGE vs MindMapQwQ-32B · ExplainCPEaccuracy (%): 84.6% -> 84.8%No clear effect
MIRAGE vs MindMap on ExplainCPE rankQwQ-32B · ExplainCPEGPT-4o rank: 3.1 vs 3.1No clear effect
swap backbone to DeepSeek-R1-32BMIRAGE · ExplainCPEaccuracy (%): 82.2% -> 84.4%Improved
add Question Decomposer (ablation)MIRAGE · GenMedGPT-5k pairwisewin rate vs ablated: win 40.72 vs lose 14.31Improved
add Answer Synthesizer with cross-chain verification (ablation)MIRAGE · GenMedGPT-5k pairwisewin rate vs ablated: win 44.03 vs lose 12.73Improved
increase max sub-questions Nq (ablation)MIRAGE · three medical QA benchmarksaccuracy and GPT-4o rankMixed
increase retrieval limit Nr (ablation)MIRAGE · three medical QA benchmarksaccuracy and GPT-4o rankImproved
QwQ-32B vs GPT-4o+ToTQwQ-32B · ExplainCPEaccuracy (%): 80.2% -> 82.8%Improved
add static BM25/dense retrieval to QwQ-32BQwQ-32B · ExplainCPEaccuracy (%): 82.8% -> 83.5% (BM25), 82.4% (dense)Mixed
Search-o1 web retrieval vs QwQ-32BQwQ-32B · ExplainCPEaccuracy (%): 82.8% -> 80.7%Hurt

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