Research map / Architectures and efficiency / Mixture of experts
DAG-MoE: From Simple Mixture to Structural Aggregation in Mixture-of-Experts
Jiarui Feng, Hanqing Zeng, Karish Grover, Ruizhong Qiu, Yinglong Xia, Qiang Zhang, Qifan Wang, Ren Chen, Dongqi Fu, Jiayi Liu, Zhuokai Zhao, Xiangjun Fan and 2 more
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 |
|---|---|---|---|
| add DAG learning module (structural aggregation) to MoE | MoE-l · Pile / Wiki / FineWeb-Edu / C4 (40B tokens) | perplexity: 10.51 -> 10.27 (Pile); 35.21 -> 34.21 (C4) | Improved |
| add DAG module instead of Chain-of-Experts iteration | CoE · Pile 12B tokens, matched params | perplexity reduction: 0.480 -> 0.587 | Improved |
| add DAG module instead of a shared expert | Standard MoE · Pile 12B tokens, matched params | perplexity reduction: 0.433 -> 0.587 | Improved |
| add structured DAG module in place of naive MLP mixing (d_g=32) (ablation) | DAG-MoE-s · Eval loss, 12B tokens | eval loss: 2.8406 -> 2.6948 | Improved |
| add naive MLP mixing of expert outputs to standard MoE (d_g=64) (ablation) | Standard MoE · Eval loss, 12B tokens | eval loss: 2.7168 -> 2.8006 | Hurt |
| increase DAG iterations from L=1 to L=2 (d_g=32) (ablation) | DAG-MoE-s · Eval loss, 12B tokens | eval loss: 2.6948 -> 2.6899 | Improved |
| increase DAG iterations L from 2 to 3 (ablation) | DAG-MoE-s/m · Pile subset, 12B tokens | perplexity | No clear effect |
| add a single DAG iteration (d_g=64, L=1) (ablation) | DAG-MoE-s · Pile subset, top-K 4 and 8 | perplexity: about 0.5 reduction in perplexity | Improved |
| use d_g=64,L=2 rather than d_g=128,L=1 (ablation) | DAG-MoE-s · Pile subset | perplexity: d_g=64,L=2 outperforms d_g=128,L=1 | Improved |
| add DAG module (L=1) vs standard MoE | DAG-MoE-s · Training throughput | tokens/s: 0.7744 -> 0.7627 | Hurt |
| add DAG module (L=2) vs standard MoE | DAG-MoE-s · Training throughput | tokens/s: 0.7744 -> 0.7404 | Hurt |
| add DAG module after instruction fine-tuning | MoE-l · GPQA | accuracy: 21.72 -> 27.78 | Improved |
Closest papers by meaning
- Advancing Expert Specialization for Better MoE (NeurIPS 2025 · similarity 0.93)
- Pruning and Distilling Mixture-of-Experts into Dense Language Models (NeurIPS 2026 · similarity 0.92)
- Scaling Laws for Fine-Grained Mixture of Experts (ICML 2024 · similarity 0.92)
- BAM! Just Like That: Simple and Efficient Parameter Upcycling for Mixture of Experts (NeurIPS 2024 · similarity 0.92)
- MoE-CAP: Benchmarking Cost, Accuracy and Performance of Sparse Mixture-of-Experts Systems (NeurIPS 2025 · similarity 0.91)
- MoEQuant: Enhancing Quantization for Mixture-of-Experts Large Language Models via Expert-Balanced Sampling and Affinity Guidance (ICML 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.
