AI Paper Atlas

Research map / Large language models / Retrieval and knowledge

Less Is More: Elevating RAG via Performance-Driven Context Compression

Ziqiang Cui, Yunpeng Weng, Xing Tang, Peiyang Liu, Shiwei Li, Bowei He, Jiamin Chen, Yansen Zhang, Xiuqiang He, Rui Zhang, Chen Ma

ICML 2026 · Large language models › Retrieval and knowledge · approach: RAG robustness and evaluation · 0 citations (OpenAlex)

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
add performance-driven RL compression instead of full top-5 documentsRAG with Qwen2.5-14B-Instruct · NQEM: 38.03 -> 41.02Improved
add performance-driven RL compression instead of full top-5 documentsRAG with Qwen2.5-14B-Instruct · 2WikiMultihopQAEM: 29.64 -> 36.72Improved
add distillation warm-start before GRPO (ablation)CORE · NQEM: 36.37 -> 41.02Improved
add RL stage after distillation (ablation)CORE · NQEM: 34.18 -> 41.02Improved
add RL stage after distillation (ablation)CORE · 2WikiMultihopQAEM: 30.25 -> 36.72Improved
add distillation warm-start before GRPO (ablation)CORE · TriviaQAEM: 65.23 -> 65.63No clear effect
add F1 reward term (alpha > 0) to EM reward (ablation)CORE reward · all four QA datasetsEMImproved
apply compressor trained on top-5 docs to top-10 docs without retrainingCORE · NQEM: 38.67 -> 41.88Improved
transfer compressor to unseen Llama-3.1-8B-Instruct readerCORE · NQEM: 37.65 -> 40.72Improved
train compressor on NQ and transfer to HotpotQACORE · HotpotQAEM: 33.67 -> 32.15Mixed
increase compressor size from 1B to 3B (ablation)CORE · QA benchmarksEMImproved
use off-the-shelf DeepSeek-V3 (671B) summaries instead of full documentsRAG with Qwen2.5-14B-Instruct · NQEM: 38.03 -> 37.73Hurt

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