混合专家论文趋势 2023–2026

主要结论

• 混合专家论文从 2023 年的 236 篇增长到 2026 年的 326 篇，三年间扩大到约 1.4 倍。
• NeurIPS 是发混合专家论文最多的会议，2023–2026 年共 354 篇，占四个会议总量的 39%。
• 交叉方向里涨得最快的是多模态大模型：同时属于两个主题的论文从 2023 年的 4 篇增至 2026 年的 79 篇。
• 在「优化与缩放」的 5 个主题里，它的 2026 年论文数排第 4；同组增长最快的缩放定律是 4.1 倍，混合专家是 1.4 倍。

代表论文

按年份和会议均匀挑选，方便快速看到每一届的样貌

• NeurIPS 2026A Biconvex Formulation for Stable Transport of Mixture Models with a Unique Solution
• ICML 2026$\phi$-Balancing for Mixture-of-Experts Training
• ICLR 2026A Federated Generalized Expectation-Maximization Algorithm for Mixture Models with an Unknown Number of Components
• CVPR 2026Divide Conquer and Aggregate Asymmetric Experts for Class-Imbalanced Semi-Supervised Medical
• NeurIPS 2025Advancing Expert Specialization for Better MoE
• ICML 2025$\texttt{I$^2$MoE}$: Interpretable Multimodal Interaction-aware Mixture-of-Experts
• ICLR 2025A Sanity Check for AI-generated Image Detection
• CVPR 2025Complexity Experts are Task-Discriminative Learners for Any Image Restoration
• NeurIPS 2024$\texttt{Model-GLUE}$: Democratized LLM Scaling for A Large Model Zoo in the Wild
• ICML 2024$\texttt{MoE-RBench}$: Towards Building Reliable Language Models with Sparse Mixture-of-Experts
• ICLR 2024Fusing Models with Complementary Expertise
• CVPR 2024MoDE: CLIP Data Experts via Clustering
• NeurIPS 2023[Re] Numerical influence of ReLU'(0) on backpropagation
• ICML 2023A Distribution Optimization Framework for Confidence Bounds of Risk Measures
• ICLR 2023Knowledge-in-Context: Towards Knowledgeable Semi-Parametric Language Models
• CVPR 2023Balanced Product of Calibrated Experts for Long-Tailed Recognition
• NeurIPS 2026AdERA: Adaptive Exponent Reuse for Lossless Allgather in Sharded MoE Training
• ICML 2026Attention Sink Forges Native MoE in Attention Layers: Sink-Aware Training to Address Head Collapse
• ICLR 2026Abstracting Robot Manipulation Skills via Mixture-of-Experts Diffusion Policies
• CVPR 2026ERMoE Eigen-Reparameterized Mixture-of-Experts for Stable Routing and Interpretable Specialization

全部混合专家论文

共 223 篇，第 3 / 5 页

101. ICML 2025M3-JEPA: Multimodal Alignment via Multi-gate MoE based on the Joint-Embedding Predictive Architecture
102. ICML 2025Make LoRA Great Again: Boosting LoRA with Adaptive Singular Values and Mixture-of-Experts Optimization Alignment
103. ICML 2025Mastering Massive Multi-Task Reinforcement Learning via Mixture-of-Expert Decision Transformer
104. ICML 2025Mastering Multiple-Expert Routing: Realizable $H$-Consistency and Strong Guarantees for Learning to Defer
105. ICML 2025MENTOR: Mixture-of-Experts Network with Task-Oriented Perturbation for Visual Reinforcement Learning
106. ICML 2025Mixture of Experts Made Intrinsically Interpretable
107. ICML 2025Mixture of Experts Provably Detect and Learn the Latent Cluster Structure in Gradient-Based Learning
108. ICML 2025Mixture of Hidden-Dimensions: Not All Hidden-States’ Dimensions are Needed in Transformer
109. ICML 2025Mixture of Lookup Experts
110. ICML 2025MoE-SVD: Structured Mixture-of-Experts LLMs Compression via Singular Value Decomposition
111. ICML 2025MoEQuant: Enhancing Quantization for Mixture-of-Experts Large Language Models via Expert-Balanced Sampling and Affinity Guidance
112. ICML 2025MoH: Multi-Head Attention as Mixture-of-Head Attention
113. ICML 2025MoHAVE: Mixture of Hierarchical Audio-Visual Experts for Robust Speech Recognition
114. ICML 2025Moirai-MoE: Empowering Time Series Foundation Models with Sparse Mixture of Experts
115. ICML 2025Multi-Modal Object Re-identification via Sparse Mixture-of-Experts
116. ICML 2025MxMoE: Mixed-precision Quantization for MoE with Accuracy and Performance Co-Design
117. ICML 2025Occult: Optimizing Collaborative Communications across Experts for Accelerated Parallel MoE Training and Inference
118. ICML 2025On Learning Parallel Pancakes with Mostly Uniform Weights
119. ICML 2025On Zero-Initialized Attention: Optimal Prompt and Gating Factor Estimation
120. ICML 2025On-Device Collaborative Language Modeling via a Mixture of Generalists and Specialists
121. ICML 2025Optimizing Robustness and Accuracy in Mixture of Experts: A Dual-Model Approach
122. ICML 2025Oracle-MoE: Locality-preserving Routing in the Oracle Space for Memory-constrained Large Language Model Inference
123. ICML 2025Organize the Web: Constructing Domains Enhances Pre-Training Data Curation
124. ICML 2025Parameters vs FLOPs: Scaling Laws for Optimal Sparsity for Mixture-of-Experts Language Models
125. ICML 2025Parrot: Multilingual Visual Instruction Tuning
126. ICML 2025Programming Every Example: Lifting Pre-training Data Quality Like Experts at Scale
127. ICML 2025QoS-Efficient Serving of Multiple Mixture-of-Expert LLMs Using Partial Runtime Reconfiguration
128. ICML 2025R2-T2: Re-Routing in Test-Time for Multimodal Mixture-of-Experts
129. ICML 2025RepLoRA: Reparameterizing Low-rank Adaptation via the Perspective of Mixture of Experts
130. ICML 2025Retraining-free Merging of Sparse MoE via Hierarchical Clustering
131. ICML 2025Scaling Laws for Upcycling Mixture-of-Experts Language Models
132. ICML 2025Score-of-Mixture Training: One-Step Generative Model Training Made Simple via Score Estimation of Mixture Distributions
133. ICML 2025Shortcut-connected Expert Parallelism for Accelerating Mixture of Experts
134. ICML 2025Soup-of-Experts: Pretraining Specialist Models via Parameters Averaging
135. ICML 2025SPACE: Your Genomic Profile Predictor is a Powerful DNA Foundation Model
136. ICML 2025TimeStep Master: Asymmetrical Mixture of Timestep LoRA Experts for Versatile and Efficient Diffusion Models in Vision
137. ICML 2025What Limits Bidirectional Model's Generative Capabilities? A Uni-Bi-Directional Mixture-of-Expert Method For Bidirectional Fine-tuning
138. NeurIPS 2025A Unified Framework for Variable Selection in Model-Based Clustering with Missing Not at Random
139. NeurIPS 2025Adaptive LoRA Experts Allocation and Selection for Federated Fine-Tuning
140. NeurIPS 2025Advancing Expert Specialization for Better MoE
141. NeurIPS 2025ALTER: All-in-One Layer Pruning and Temporal Expert Routing for Efficient Diffusion Generation
142. NeurIPS 2025Blindfolded Experts Generalize Better: Insights from Robotic Manipulation and Videogames
143. NeurIPS 2025Brain-Like Processing Pathways Form in Models With Heterogeneous Experts
144. NeurIPS 2025BrainMoE: Cognition Joint Embedding via Mixture-of-Expert Towards Robust Brain Foundation Model
145. NeurIPS 2025Consistent Sampling and Simulation: Molecular Dynamics with Energy-Based Diffusion Models
146. NeurIPS 2025Continual Gaussian Mixture Distribution Modeling for Class Incremental Semantic Segmentation
147. NeurIPS 2025CryptoMoE: Privacy-Preserving and Scalable Mixture of Experts Inference via Balanced Expert Routing
148. NeurIPS 2025Data Mixture Optimization: A Multi-fidelity Multi-scale Bayesian Framework
149. NeurIPS 2025Deferring Concept Bottleneck Models: Learning to Defer Interventions to Inaccurate Experts
150. NeurIPS 2025Dense Backpropagation Improves Training for Sparse Mixture-of-Experts

常见问题

2023–2026 年四大会议一共有多少篇混合专家论文？

共 904 篇，按年份分别是 2023 年 236 篇、2024 年 119 篇、2025 年 223 篇、2026 年 326 篇。

NeurIPS 2026 有多少篇混合专家论文？

NeurIPS 2026 有 75 篇混合专家论文，上一届是 86 篇。

ICLR 2026 有多少篇混合专家论文？

ICLR 2026 有 73 篇混合专家论文，上一届是 59 篇。

ICML 2026 有多少篇混合专家论文？

ICML 2026 有 132 篇混合专家论文，上一届是 57 篇。

CVPR 2026 有多少篇混合专家论文？

CVPR 2026 有 46 篇混合专家论文，上一届是 21 篇。

哪个会议的混合专家论文最多？

NeurIPS，2023–2026 年共 354 篇。