AI Paper Atlas

Research map / Architectures and efficiency / Mixture of experts

MoE-GRPO: Optimizing Mixture-of-Experts via Reinforcement Learning in Vision-Language Models

Dohwan Ko, Jinyoung Park, Seoung Choi, Sanghyeok Lee, Seohyun Lee, Hyunwoo J. Kim

CVPR 2026 · Architectures and efficiency › Mixture of experts · approach: Expert routing and specialization · 4 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 deterministic top-K fine-tuning with MoE-GRPOInternVL3.5 MoE · 8 multimodal benchmarks avgaccuracy (%): 54.0 -> 56.0Improved
add stochastic expert sampling (multinomial or noise) without RLInternVL3.5 MoE · 8 multimodal benchmarks avgaccuracy (%): 54.0 -> 53.7 (multi), 54.3 (noise)No clear effect
apply MoE-GRPO instead of Det-FT on CLIP-MoECLIP-MoE · Cross-dataset eval avg (10 datasets)accuracy (%): 68.8 -> 71.9Improved
fine-tune CLIP-MoE with Det-FTCLIP-MoE · Cross-dataset eval avgaccuracy (%): 70.8 -> 68.8Hurt
apply MoE-GRPO instead of Det-FT for domain generalizationCLIP-MoE · ImageNet variants avgaccuracy (%): 66.0 -> 67.5Improved
add Gate-GRPO to Token-GRPO (ablation)MoE-GRPO · 4-benchmark ablation avgaccuracy (%): 53.9 -> 55.7Improved
add Token-GRPO to Gate-GRPO (ablation)MoE-GRPO · 4-benchmark ablation avgaccuracy (%): 50.9 -> 55.7Improved
use modality-aware router guidance instead of Gaussian-noise exploration (ablation)MoE-GRPO · 4-benchmark ablation avgaccuracy (%): 54.2 -> 55.7Improved
use modality-aware router guidance instead of multinomial exploration (ablation)MoE-GRPO · 4-benchmark ablation avgaccuracy (%): 54.8 -> 55.7Improved
swap GRPO for DAPO or SAPO (ablation)MoE-GRPO · 4-benchmark ablation avgaccuracy (%): 55.7 -> 55.0 (DAPO), 55.4 (SAPO)No clear effect
use MoE-GRPO instead of Expert Choice or Optimal Transport routing (ablation)MoE · 4-benchmark ablation avgaccuracy (%): 54.7 / 54.4 -> 55.7Improved
add load-balancing loss to MoE-GRPO (ablation)MoE-GRPO · 4-benchmark ablation avgaccuracy (%): 54.8 -> 55.7Improved

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