AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

RCMoE: A Communication-Efficient Random Compression Framework for Resource-Constrained Mixture-of-Experts Training

Donglei Wu, Xiao Cai, Jinglei Tan, Jinda Jia, Guangming Tan, Dingwen Tao, Wen Xia, Zhihong Tian

AAAI 2026 · Architectures and efficiency › Mixture of experts · approach: Expert routing and specialization · 1 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
row-wise quantization instead of column-wise or global quantization of intermediate results (8-bit) (ablation)MoE all-to-all compression · MoE training validation accuracyvalidation accuracyImproved
row-wise instead of global quantization (ablation)MoE all-to-all compression · MoE training runtimeruntimeImproved
increase quantization bit-width from 6 to 8 bits (ablation)linear quantization of intermediate results · MoE trainingaccuracy vs compression ratioMixed
use unbiased LSQ instead of biased global linear quantization (ShapeShifter) (ablation)all-to-all compression · VC100convergence stabilityImproved
use LSQ instead of QSGD for intermediate results (ablation)all-to-all compression · MoE taskstraining convergenceImproved
use PTS instead of MSTop-K or Rand-K sparsification (ablation)all-reduce compression · MoE tasks at same sparsitytraining qualityImproved
use PTS instead of DGC sparsification (ablation)all-reduce compression · ViT-based vision taskstraining qualityImproved
RCMoE (LSQ+PTS) vs best compression combinationsGW2 · GW2 validation perplexityperplexity: 311.1 -> 304.5Improved
RCMoE vs no compressionVT200 · VT200 validation accuracyaccuracy: 0.5324 -> 0.5861Improved
naive compression combinations (Top-k, linear quantization) applied to MoE trainingMoE training · MoE tasksaccuracy: 18.9%-66.7% accuracy degradationHurt
PWLQ/ShapeShifter+DGC combos in speedupMoE training time · VT200speedup: 0.81xHurt
scale to more GPUs with RCMoERCMoE · 4 to 16 GPUsspeedupImproved

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