AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

MergePRAG: Orthogonal Merging of Passage-experts for Multi-hop Parametric RAG

Xuebing Liu, Shanbao Qiao, Roseline Nyange, Dongwook Min, Hyun Kim, Seung-Hoon Na

ICLR 2026 · Large language models › Retrieval and knowledge · approach: RAG robustness and evaluation

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 orthogonal (Gram-Schmidt) sequence merging instead of TIES merging (ablation)MergePRAG+ |SP|=1 · HotpotQAEM: 46.40 -> 48.80Improved
use orthogonal sequence merging instead of additive merging (ablation)MergePRAG+ |SP|=1 · HotpotQAEM: 36.20 -> 48.80Improved
use orthogonal sequence merging instead of arithmetic mean (ablation)MergePRAG+ |SP|=1 · HotpotQAEM: 48.20 -> 48.80No clear effect
use orthogonal sequence merging instead of concat merging (ablation)MergePRAG+ |SP|=1 · HotpotQAEM: 48.20 -> 48.80No clear effect
use orthogonal merging for inner and sequence merging instead of arithmetic mean (ablation)MergePRAG+ |SP|=4 · HotpotQAEM: 51.40 -> 52.40Improved
use hypernetwork parameterization (MergePRAG+) instead of finetuning [sq->sa]MultihopRAG LLaMA3.1-8B · HotpotQAEM: 43.70 -> 47.40Improved
finetune with gold passages instead of without passagesMultihopRAG LLaMA3.1-8B · HotpotQAEM: 43.70 -> 40.10Hurt
add hypernetwork passage injection to MultihopRAGMultihopRAG LLaMA3.1-8B · HotpotQAEM: 37.80 -> 47.40Improved
add in-context passages to MergePRAG (MergePRAG+) (ablation)MergePRAG LLaMA3.1-8B · HotpotQAEM: 28.40 -> 48.80Improved
increase retrieved passages per hop from 1 to 4 (ablation)MergePRAG+ · HotpotQAEM: 48.80 -> 52.40Improved
increase number of key-value vectors k from 1 to 16 (ablation)MergePRAG+ |SP|=4 · HotpotQAEM: 49.00 -> 52.40Improved
compare MergePRAG+ against R3-RAG on a harder datasetMergePRAG+ |SP|=4 LLaMA3.1-8B · MuSiQue (E5)EM: 21.20 -> 16.70Hurt

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