AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

Routing Manifold Alignment Improves Generalization of Mixture-of-Experts LLMs

Zhongyang Li, Ziyue Li, Tianyi Zhou

ICLR 2026 · Architectures and efficiency › Mixture of experts · approach: Efficient MoE language model systems · 12 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 routing manifold regularization to router tuning (RoMA)OLMoE-7B-A1B · MMLUaccuracy: 57.8 -> 69.0Improved
add RoMADeepSeekMoE-16B-A3B · MMLUaccuracy: 46.2 -> 56.8Improved
add RoMAQwen3-30B-A3B · MMLUaccuracy: 74.2 -> 78.8Improved
add RoMA to OLMoE on out-of-distribution mathOLMoE-7B-A1B · GSM8Kaccuracy: 45.5 -> 49.4Improved
RoMA instead of C3PO test-time routingOLMoE-7B-A1B · 8-benchmark averageaccuracy: 75.7 -> 76.2No clear effect
regularize only last five router layers instead of all (ablation)RoMA on OLMoE · 8-benchmark averageaccuracy: 75.1 -> 76.2Improved
regularize last token only rather than first or middle token (ablation)RoMA on OLMoE · 8-benchmark averageaccuracy: 71.4 / 69.2 -> 76.2Improved
use k-NN successful neighbors instead of random neighbors (ablation)RoMA on OLMoE · 8-benchmark averageaccuracy: 67.8 -> 76.2Improved
use manifold regularization instead of L1, L2, or entropy (ablation)RoMA on OLMoE · 8-benchmark averageaccuracy: 71.5 -> 76.2Improved
increase k from 1 or 5 to 3 in k-NN neighbors (ablation)RoMA on OLMoE · 8-benchmark averageaccuracyImproved
increase training data from 10% to 100% (ablation)RoMA on OLMoE · 8-benchmark averageaccuracy: 68.5 -> 76.2Improved
increase epsilon-ball radius beyond 0.5 to 0.7 (ablation)RoMA on OLMoE · 8-benchmark averageaccuracyHurt

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