AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Learning to Route Languages for Multilingual Policy Optimization

Geyang Guo, Hiromi Wakaki, Yuki Mitsufuji, Alan Ritter, Wei Xu

ICML 2026 · Reinforcement learning › RL from feedback · approach: GRPO stability and variants · 0 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 GRPO with LRPO (routed multilingual rollouts)Qwen2.5-1.5B-it · mGSM-v2 averagescore: 32.33 -> 38.25Improved
replace GRPO with LRPOQwen2.5-1.5B-it · Overall average (5 benchmarks)average score: 30.42 -> 32.15Improved
replace GRPO with LRPOLlama3.2-1B-it · mGSM-v2 averagescore: 21.24 -> 20.55Hurt
replace GRPO with LRPOGemma3-4B-it · CARE-pro averagescore: 18.92 -> 18.45Hurt
use multilingual rollouts instead of monolingual (ablation)LRPO fixed router · Overall average (Qwen2.5-1.5B-it)average score: 30.42 -> 31.89Improved
sample languages uniformly instead of EN-dominant mix (ablation)LRPO fixed router · Overall average (Qwen2.5-1.5B-it)average score: 31.89 -> 31.36Hurt
use dynamic learned router instead of fixed router (ablation)LRPO · Overall average (Qwen2.5-1.5B-it)average score: 31.89 -> 32.15Improved
add uniform initialization of the router (ablation)LRPO router · Overall average (Qwen2.5-1.5B-it)average score: 31.48 -> 32.15Improved
add temperature annealing to the router (ablation)LRPO router · Overall average (Qwen2.5-1.5B-it)average score: 31.67 -> 32.15Mixed
add warm-start SFT for language control (ablation)LRPO · Overall average (Qwen2.5-1.5B-it)average score: 30.44 -> 32.15Improved
LRPO without warm-start vs GRPO (ablation)LRPO-Zero · Overall average (Qwen2.5-1.5B-it)average score: 30.42 -> 30.44No clear effect
add LRPO on multiple-choice tasksLRPO · Global-MMLU-Lite / Include-LiteaccuracyNo 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