Research map / Architectures and efficiency / Mixture of experts
ERMoE: Eigen-Reparameterized Mixture-of-Experts for Stable Routing and Interpretable Specialization
Anzhe Cheng, Shukai Duan, Shixuan Li, Chenzhong Yin, Mingxi Cheng, Heng Ping, Tamoghna Chattopadhyay, Sophia I. Thomopoulos, Shahin Nazarian, Paul Thompson, Paul Bogdan
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 |
|---|---|---|---|
| replace V-MoE routing with ERMoE eigenbasis routing (no load-balancing loss) | ViT-B/16 · ImageNet-1K val | Top-1 accuracy (%): 87.41 -> 88.03 | Improved |
| use ERMoE instead of V-MoE | ViT-B/16 linear probe · CIFAR-100 (5-shot) | Top-1 accuracy (%): 89.49 -> 93.47 | Improved |
| use ERMoE instead of V-MoE | ViT-B/16 linear probe · Tiny-ImageNet-200 (5-shot) | Top-1 accuracy (%): 81.12 -> 83.17 | Improved |
| use ERMoE instead of V-MoE | ViT-B/16 linear probe · CIFAR-10 (10-shot) | Top-1 accuracy (%): 98.94 -> 98.79 | No clear effect |
| use CLIP-ERMoE image encoder instead of CLIP-MoE | CLIP · MS COCO I2T | R@10: 92.0 -> 94.7 | Improved |
| use CLIP-ERMoE image encoder instead of Fix-CLIP | CLIP · MS COCO T2I | R@10: 82.4 -> 81.6 | Hurt |
| use CLIP-ERMoE instead of CLIP-MoE | CLIP · Flickr30k T2I | R@1: 42.1 -> 42.4 | No clear effect |
| use ERMoE-ba instead of 3D Swin Transformer | 3D ViT · ADNI brain age | MAE (years): 2.83 -> 2.31 | Improved |
| use ERMoE-ba instead of 3D CNN | 3D ViT · ADNI brain age | MAE (years): 3.13 -> 2.31 | Improved |
Closest papers by meaning
- Multilinear Mixture of Experts: Scalable Expert Specialization through Factorization (NeurIPS 2024 · similarity 0.89)
- A Provably Effective Method for Pruning Experts in Fine-tuned Sparse Mixture-of-Experts (ICML 2024 · similarity 0.88)
- From Sparse to Soft Mixtures of Experts (ICLR 2024 · similarity 0.88)
- Tight Clusters Make Specialized Experts (ICLR 2025 · similarity 0.87)
- Efficient Data Driven Mixture-of-Expert Extraction from Trained Networks (CVPR 2025 · similarity 0.87)
- Dense Backpropagation Improves Training for Sparse Mixture-of-Experts (NeurIPS 2025 · similarity 0.87)
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.
