Research map / Architectures and efficiency / Mixture of experts
Towards Principled Fine-Grained MoE Expert Pruning via Pseudo-Boolean Approximation
Zongfang Liu, Ziheng Cheng, Shengkun Tang, Jinghui Zhang, Weijie Wang, Zhiqiang Shen, Xin Yuan
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a principled fine-grained expert pruning method for MoE language models that uses a pseudo-Boolean approximation to balance expert interaction modeling with search tractability.
Problem
Storing all experts of MoE language models at inference makes memory a deployment bottleneck; existing expert pruning methods face a tension between search-based approaches that capture expert interactions and cheaper approaches that ignore them.
Approach
Formulates fine-grained expert pruning as a pseudo-Boolean optimization problem and uses a pseudo-Boolean approximation to capture expert interaction effects while keeping the search tractable.
Closest papers by meaning
- FlexMoE: One-for-All Nested Intra-Expert Pruning for MoE Language Models (NeurIPS 2026 · similarity 0.93)
- Pruning and Distilling Mixture-of-Experts into Dense Language Models (NeurIPS 2026 · similarity 0.93)
- Domain-Specific Pruning of Large Mixture-of-Experts Models with Few-shot Demonstrations (NeurIPS 2025 · similarity 0.92)
- DiEP: Adaptive Mixture-of-Experts Compression through Differentiable Expert Pruning (NeurIPS 2025 · similarity 0.91)
- A Provably Effective Method for Pruning Experts in Fine-tuned Sparse Mixture-of-Experts (ICML 2024 · similarity 0.91)
- Advancing Expert Specialization for Better MoE (NeurIPS 2025 · similarity 0.91)
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.
