AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

COMI: Coarse-to-fine Context Compression via Marginal Information Gain

Jiwei Tang, Shilei Liu, Zhicheng Zhang, Yujin Yuan, Libin Zheng, Wenbo Su, Bo Zheng

ICLR 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 11 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
replace uniform grouping with MIG-based group reallocation (ablation)COMI · NaturalQuestions 32x Qwen2-7BEM: 54.54 -> 56.31Improved
replace average pooling with MIG-weighted token merging (ablation)COMI · NaturalQuestions 32x Qwen2-7BEM: 50.81 -> 56.31Improved
add inter-group redundancy to relevance in reallocation (ablation)COMI · NaturalQuestions 32x Qwen2-7BEM: 54.59 -> 56.31Improved
add intra-group redundancy penalty in token merging (ablation)COMI · NaturalQuestions 32x Qwen2-7BEM: 51.53 -> 56.31Improved
add MIG reallocation on 2WikiMQA (ablation)COMI · 2WikiMQA 32x Qwen2-7BEM: 49.82 -> 52.13Improved
use MIG instead of relevance for answer-segment prediction (ablation)MIG · NaturalQuestionsAUC: 0.5423 -> 0.5809Improved
use MIG instead of relevance at 25% retention (ablation)MIG · NaturalQuestionsredundancy score (lower better): 0.6839 -> 0.5673Improved
COMI vs best baseline at 32x on NaturalQuestionsCOMI · NaturalQuestions 32x Qwen2-7BEM: 24.63 -> 49.15Improved
COMI vs original prompt with native long-context modelCOMI · NaturalQuestions 16x Qwen3-4BF1: 16.90 -> 34.79Improved
COMI vs original prompt on strong Qwen2 model at 32xCOMI · NaturalQuestions Qwen2-7BEM: 72.35 -> 49.15Hurt
COMI vs Activation Beacon on latencyCOMI · NarrativeQA 32xend-to-end latency (s): 3.93 -> 3.27Mixed

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