AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

Expert Divergence Learning for MoE-based Language Models

Jiaang Li, Haibin Chen, Langming Liu, Yujin Yuan, Yadao Wang, Yizhen Zhang, Chengting Yu, Xin Tong, Weidong Zhang, Shilei Liu, Wenbo Su, Bo Zheng

ICLR 2026 · Architectures and efficiency › Mixture of experts · approach: Expert routing and specialization · 3 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 49-class expert divergence lossMoE 15B-A1.5B · 7-benchmark averageaverage accuracy: 35.59 -> 36.65Improved
add 3-class expert divergence lossMoE 15B-A1.5B · 7-benchmark averageaverage accuracy: 35.59 -> 36.34Improved
add 49-class expert divergence lossMoE 8B-A0.8B · 7-benchmark averageaverage accuracy: 35.10 -> 35.77Improved
add 3-class expert divergence lossMoE 8B-A0.8B · 7-benchmark averageaverage accuracy: 35.10 -> 35.54Improved
add 49-class expert divergence lossMoE 3B-A0.3B · 7-benchmark averageaverage accuracy: 34.13 -> 34.91Improved
add 3-class expert divergence lossMoE 3B-A0.3B · 7-benchmark averageaverage accuracy: 34.13 -> 33.98No clear effect
use 49-class instead of 3-class domain labels (ablation)expert divergence loss · 15B-A1.5B averageaverage accuracy: 36.34 -> 36.65Improved
use 49-class instead of 3-class domain labels (ablation)expert divergence loss · 3B-A0.3B averageaverage accuracy: 33.98 -> 34.91Improved
treat each document as its own domain instead of 49 classes (ablation)expert divergence loss · 3B training lossLM lossHurt
add expert divergence lossMoE 15B-A1.5B · Training throughputk tokens/s/gpu: 5.07 -> 4.96No clear effect
add expert divergence lossMoE 15B-A1.5B · Inference throughput (en_val_1000)tokens/s: 14.45 -> 14.39No clear effect
vary divergence coefficient betaexpert divergence loss · 3B-A0.3B training lossLM lossNo clear effect

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