AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

MRACL: Multi-Reward Space Guided Adaptive Curriculum Reinforcement Learning for LLMs

Wenxuan Liu, Liangyu Huo, Yi Jing, Xiyuan Zhang, Jian Xie

AAAI 2026 · Reinforcement learning › RL from feedback · approach: RL for LLM reasoning · 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
use MRACL multi-reward curriculum instead of DUMPReinforce++ on Qwen2.5-7B-Base · UltraFeedback in-distributionconvergence speed: 1.62x faster than DUMPImproved
use MRACL curriculum instead of no curriculumReinforce++ on Qwen2.5-7B-Base · UltraFeedback in-distributionconvergence speed: 2.55x faster (abstract) vs 2.54x (results text)Improved
use MRACL instead of DUMPQwen2.5-7B-Base · win rate (step 100 avg)win rate (%): 60.0 -> 63.1Improved
use MRACL instead of the best baseline on rule-based benchmarksQwen2.5-7B-Base · TriviaQAscore: 51.71 -> 50.31 (SpeedRL vs MRACL)Hurt
add capability distance term (ablation)MRACL · UltraFeedback hard subsetsum rewardImproved
add optimization direction term (ablation)MRACL · UltraFeedback hard subsetsum reward: 25.0 -> 26.6Improved
add historical evolution term (ablation)MRACL · UltraFeedback easy/median/hardsum rewardImproved
add exploration (UCB bias) (ablation)MRACL · UltraFeedbacksum reward (later training)Improved
use fixed alpha=beta=0.5 instead of time-varying schedule (ablation)MRACL · UltraFeedbackfinal sum rewardHurt
reverse the alpha/beta schedule (ablation)MRACL · UltraFeedbacksum rewardHurt
reduce rollouts from 4 to 1 (ablation)DUMP · UltraFeedback sum rewardfinal sum reward: 33.8 -> 32.1Hurt
apply MRACL to a stronger instruct modelQwen2.5-7B-Instruct · UltraFeedback easy/median vs hardsum reward: hard subset 25.1 -> 30Mixed

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