AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

ConceptMoE: Adaptive Token-to-Concept Compression for Implicit Compute Allocation

Zihao Huang, Jundong Zhou, Xingwei Qu, Qiyang Min, Ge Zhang

ICML 2026 · Architectures and efficiency › Attention and transformers · approach: Sparse attention for long-context LLMs · 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
use ConceptMoE instead of MoEMoE-A0.5B-12B · OpenBench EasyAll score: 35.6 -> 36.4Improved
use ConceptMoE instead of MoEMoE-A1B-24B · OpenBench EasyAll score: 50.0 -> 50.9Improved
use ConceptMoE instead of MoEMoE-A2.5B-60B · Long contextAll score: 49.4 -> 51.7Improved
use ConceptMoE instead of MoEMoE-A2.5B-60B · Vision-languageAll score: 53.6 -> 54.2Improved
convert MoE to ConceptMoE-top15 in continual trainingMoE-A2.5B-90B · Open BenchmarkOverall score: 40.9 -> 41.3No clear effect
add layer loops (ConceptMoE-top11-loop8) in CT conversionMoE-A2.5B-90B · Open BenchmarkOverall score: 40.9 -> 46.4Improved
train ConceptMoE-top11-loop8 from scratch rather than CT conversionConceptMoE-top11-loop8 · Open BenchmarkOverall score: 46.4 -> 47.3Improved
add learnable dynamic chunking instead of fixed chunking (ablation)ConceptMoE · OpenBench Easy All (A0.5B-12B)All score: 34.2 -> 36.4Improved
use cosine router instead of linear router (ablation)ConceptMoE · OpenBench Easy All (A0.5B-12B)All score: 34.4 -> 36.4Improved
use cosine router instead of linear router (ablation)ConceptMoE · A0.5B-12Btraining loss: 0.003 lower loss for linearHurt
add joint decoding of concepts and tokens (ablation)ConceptMoE · OpenBench Easy All (A0.5B-12B)All score: 35.1 -> 36.4Improved
add Bernoulli boundary noise tau=4 (ablation)ConceptMoE · OpenBench Easy All (R=1.5)All score: 28.9 -> 30.3Improved

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