Research map / Architectures and efficiency / Mixture of experts
How Many Experts Are Enough? Towards Optimal Semantic Specialization for Mixture-of-Experts
Sumin Park, Noseong Park
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 changed | Where | Measured | Effect |
|---|---|---|---|
| use MASS adaptive expansion with top-p routing instead of fixed-size MoE | Naive MoE · Synthetic HMM | test loss: 12.4 experts avg, loss 2.15 near elbow K=15 | Improved |
| use MASS instead of fixed top-k routing | Naive MoE top-k · Synthetic HMM | active experts per token: 3.9 active experts, comparable or better loss | Improved |
| use MASS instead of Naive MoE for expert specialization | Naive MoE · Synthetic HMM | routing JSD | Improved |
| use MASS instead of DynMoE | DynMoE · GLUE (5 tasks) | top-1 accuracy / active experts: active k 6.5-8.0 -> 2.6-3.2 | Improved |
| use MASS instead of DynMoE | DynMoE · DomainBed PACS | accuracy: 88.4 -> 88.7 | Improved |
| use MASS instead of DynMoE | DynMoE · DomainBed VLCS | accuracy: 79.4 -> 81.1 | Improved |
| use MASS instead of DynMoE | DynMoE · DomainBed TerraIncognita | accuracy: 46.63 -> 47.5 | Improved |
| use MASS instead of best fixed GMoE | GMoE (K=6) · DomainBed average | accuracy: 72.8 -> 72.8 | No clear effect |
| use MASS instead of GMoE K=6 | GMoE (K=6) · DomainBed OfficeHome | accuracy: 74.2 -> 73.8 | Hurt |
| use MASS instead of GMoE K=6 | GMoE (K=6) · DomainBed TerraIncognita | accuracy: 48.5 -> 47.5 | Hurt |
Closest papers by meaning
- Advancing Expert Specialization for Better MoE (NeurIPS 2025 · similarity 0.91)
- Tight Clusters Make Specialized Experts (ICLR 2025 · similarity 0.89)
- Autonomy-of-Experts Models (ICML 2025 · similarity 0.89)
- Layerwise Recurrent Router for Mixture-of-Experts (ICLR 2025 · similarity 0.88)
- Routing Experts: Learning to Route Dynamic Experts in Existing Multi-modal Large Language Models (ICLR 2025 · similarity 0.88)
- Domain-Specific Pruning of Large Mixture-of-Experts Models with Few-shot Demonstrations (NeurIPS 2025 · similarity 0.88)
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.
