AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

Routing Matters in MoE: Scaling Diffusion Transformers with Explicit Routing Guidance

Yujie Wei, Shiwei Zhang, Hangjie Yuan, Yujin Han, Zhekai Chen, Jiayu Wang, Difan Zou, Xihui Liu, Yingya Zhang, Yu Liu, Hongming Shan

ICLR 2026 · Architectures and efficiency › Mixture of experts · approach: Efficient MoE language model systems

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 changedWhereMeasuredEffect
use ProMoE two-step prototypical MoE instead of dense DiT at equal activated parametersDiT-B/L/XL, Rectified Flow · ImageNet 256x256, 500K stepsFID50K (CFG=1.5): 9.02 -> 6.39 (B), 3.56 -> 2.79 (L), 3.23 -> 2.59 (XL)Improved
use ProMoE under the DDPM objectiveDiT-B DDPM · ImageNet 256x256, 500K stepsFID50K (CFG=1.5): 18.61 -> 17.90Improved
existing DiT-MoE token-choice routing instead of denseDiT-MoE-B-Flow · ImageNet 256x256, 500K steps, CFG=1.5FID50K: 9.02 -> 8.94No clear effect
add semantic routing guidance via classification loss or k-means routing (ablation)MoE DiT-B-Flow · ImageNet 256x256, 500K steps, CFG=1.5FID50K: 8.22 (DiffMoE) -> 5.91 (classification), 6.24 (k-means)Improved
add prototypical routing (ablation)Dense-DiT-B-Flow · ImageNet 256x256, 500K stepsFID50K (CFG=1.5): 9.02 -> 7.92Improved
add routing contrastive learning (ablation)ProMoE-B-Flow · ImageNet 256x256, 500K stepsFID50K (CFG=1.5): 7.92 -> 6.75Improved
add conditional routing with unconditional experts (ablation)ProMoE-B-Flow · ImageNet 256x256, 500K stepsFID50K (CFG=1.5): 6.75 -> 6.39Improved
add conditional routing to K-means routing (ablation)K-means-based routing · ImageNet 256x256, 500K stepsFID50K (CFG=1.5): 8.75 -> 6.24Improved
use identity score activation instead of softmax or sigmoid (ablation)prototypical routing · ImageNet 256x256, 500K stepsFID50K (CFG=1.5): 6.92 (softmax) -> 6.39 (identity)Improved
add conventional load-balancing loss on top of RCL (ablation)ProMoE-B-Flow · ImageNet 256x256, 500K stepsFID50K (CFG=1.5): 6.39 -> 6.53Hurt
raise RCL loss weight from 1 to 10 (ablation)ProMoE-B-DDPM · ImageNet 256x256, 500K stepsFID50K (CFG=1.5): 18.34 -> 17.90No clear effect
increase activated routed experts from 1 to 3 (ablation)ProMoE-L-Flow · ImageNet 256x256, 500K stepsFID50K (CFG=1.5): 2.79 -> 2.72Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy