AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

Rewiring Experts on the Fly: Continuous Rerouting for Better Online Adaptation in Mixture-of-Expert Models

Guinan Su, Yanwu Yang, Li Shen, Lu Yin, Shiwei Liu, Jonas Geiping

ICML 2026 · Architectures and efficiency › Mixture of experts · approach: Expert routing and specialization

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 test-time router rerouting with confidence-selected layersDeepSeek-V2-Lite · Five-benchmark averageaverage accuracy: 50.89 -> 53.59Improved
add C3PO reference-based rerouting (100 references)DeepSeek-V2-Lite · Five-benchmark averageaverage accuracy: 50.89 -> 48.10Hurt
add 3-shot or 5-shot in-context learningDeepSeek-V2-Lite · Five-benchmark averageaverage accuracy: 50.89 -> 49.26 (3-shot), 49.08 (5-shot)Hurt
select layers by router confidence instead of randomly (ablation)Rewiring · Five-benchmark averageaverage accuracy: 49.77 -> 53.59Improved
update only high-confidence layers instead of all layers (ablation)Rewiring · Five-benchmark averageaverage accuracy: 50.68 -> 53.59Improved
add continuous re-optimization every 128 generated tokens (ablation)Rewiring · Five-benchmark averageaverage accuracy: 52.55 -> 53.59Improved
add rewiring to 3-shot ICLICL (3-shot) · Five-benchmark averageaverage accuracy: 49.26 -> 53.05Improved
add rewiring to self-consistency (3 paths)Self-Consistency (3) · Five-benchmark averageaverage accuracy: 54.88 -> 57.09Improved
add rewiring to GPT-OSS-20BGPT-OSS-20B · AIME24average correctness (%): 69.58 -> 73.75Improved
add rewiring to OLMoEOLMoE · MMLU-reduxaccuracy: 37.17 -> 38.30No clear effect
add rewiring to sample generationDeepSeek-V2-Lite · HumanEval 10 samplescosine diversity: 0.390 -> 0.380No clear effect
re-optimize more frequently (every 96 tokens) or never (ablation)Rewiring · Five benchmarksaccuracyHurt

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