Research map / Architectures and efficiency / Mixture of experts
The Expert Strikes Back: Interpreting Mixture-of-Experts Language Models at Expert Level
Jeremy Herbst, Stefan Wermter, Jae Hee Lee
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 |
|---|---|---|---|
| increase clustering resolution k from 10 to 5000 | OLMoE-L15-E17 · Functional specialization | JSD: 0.049 -> 0.349 | Improved |
| increase clustering resolution k from 10 to 5000 | OLMoE-L14-E59 · Functional specialization | JSD: 0.106 -> 0.417 | Improved |
| increase clustering resolution k from 10 to 5000 | OLMoE-L9-E60 · Functional specialization | JSD: 0.033 -> 0.130 | Improved |
| increase clustering resolution k from 10 to 5000 | OLMoE-L15-E17 · Routing specialization | JSD: 0.012 -> 0.065 | No clear effect |
| increase clustering resolution k from 10 to 5000 | OLMoE-L9-E60 · Routing specialization | JSD: 0.112 -> 0.176 | Improved |
| use sparser routing (Mixtral N_A/N=0.25 to Qwen3-30B-A3B N_A/N~0.06) | MoE models · k-sparse probing F1 | F1: 0.25 -> 0.06 | Improved |
| use a sparse MoE (OLMoE-1B-7B) instead of dense OLMo-7B | OLMoE-1B-7B · k-sparse probing | F1 | Improved |
| use MoE experts instead of dense FFNs | MoE models · k-sparse probing (k=1) | F1 | Improved |
| move from matched prompts to control prompts for the same expert | OLMoE-1B-7B experts · Causal attribution | expert routed (%): 80% | Improved |
| use experts with sparser routing for automatic labeling | Qwen3-30B-A3B · Automatic interpretability | scorer F1 | Improved |
Closest papers by meaning
- Mixture of Experts Made Intrinsically Interpretable (ICML 2025 · similarity 0.95)
- Mixture-of-Experts Meets Instruction Tuning: A Winning Combination for Large Language Models (ICLR 2024 · similarity 0.92)
- Advancing Expert Specialization for Better MoE (NeurIPS 2025 · similarity 0.92)
- Pruning and Distilling Mixture-of-Experts into Dense Language Models (NeurIPS 2026 · similarity 0.91)
- Parameters vs FLOPs: Scaling Laws for Optimal Sparsity for Mixture-of-Experts Language Models (ICML 2025 · similarity 0.90)
- BAM! Just Like That: Simple and Efficient Parameter Upcycling for Mixture of Experts (NeurIPS 2024 · similarity 0.90)
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.
