AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

Spatial Mixture-of-Experts

Nikoli Dryden, Torsten Hoefler

NeurIPS 2022 · Architectures and efficiency › Mixture of experts · approach: Sparse MoE as regularization and attention · 5 citations (OpenAlex)

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
Replace all 3×3 convolutions with SMoE layers and add a land-sea gate priorResNet on WeatherBench · WeatherBenchlatitude-weighted RMSE: 316 -> 270; 563 -> 525; 1.80 -> 1.66; 2.84 -> 2.60Improved
Replace U-Net 3×3 convolutions with SMoE layersU-Net · ENS-10CRPS: 76.54 -> 68.94; 0.685 -> 0.612; 0.657 -> 0.601Improved
Replace all 3×3 convolutions with SMoE layersResNet-50 · ImageNet-1k valTop-1 accuracy: 80.83 -> 81.33Improved
Add routing classification loss (ablation)SMoE · Heat diffusion% within 1%: 91.5 -> 96.7Improved
Add expert error damping to SMoE with routing classification loss (ablation)SMoE · Heat diffusion% within 1%: 96.7 -> 100.0Improved
Add expert error damping without routing classification loss (ablation)SMoE · Heat diffusion% within 1%: 91.5 -> 91.6No clear effect
Use tensor routing instead of a fully-connected, convolutional, CoordConv, or LCN gate (ablation)SMoE · Heat diffusion% within 1%: 85.2 -> 100.0; 91.3 -> 100.0; 91.5 -> 100.0; 96.3 -> 100.0Improved
Add a land-sea prior to the learned gate (ablation)SMoE · WeatherBenchlatitude-weighted RMSE: 278 -> 270; 530 -> 525; 1.69 -> 1.66; 2.65 -> 2.60Improved
Fix a randomly initialized gate instead of learning the routing (ablation)SMoE · WeatherBenchlatitude-weighted RMSE: 278 -> 328; 530 -> 572; 1.69 -> 1.89; 2.65 -> 2.96Hurt
Add routing noise (ablation)SMoE with RC loss and damping · Heat diffusion% within 1%: 100.0 -> 95.7Hurt
Replace convolution experts with CoordConv experts (ablation)SMoE · Heat diffusion% within 1%: 100.0 -> 96.5Hurt
Remove gate-based expert-output weighting (ablation)SMoE · Heat diffusion% within 1%: 100.0 -> 100.0No clear effect

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