Research map / Reinforcement learning / RL from feedback
MetaAct-RL: Training Language Models for Reasoning Through Meta-Action-Based Reinforcement Learning
Zhiheng Xi, Yuhui Wang, Yiwen Ding, Guanyu Li, Senjie Jin, Shichun Liu, Jixuan Huang, Dingwen Yang, Jiafu Tang, Boyang Hong, Junjie Ye, Shihan Dou and 7 more
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add length-based reward (ablation) | MetaAct-RL (Llama3.2-1B) · AQuA | accuracy: 43.70 -> 48.43 | Improved |
| add key-state restart (ablation) | MetaAct-RL (Llama3.2-1B) · AQuA | accuracy: 44.49 -> 48.43 | Improved |
| restart from forward-reasoning step instead of key critique state (ablation) | MetaAct-RL (Llama3.2-1B) · MATH | accuracy: 25.00 -> 23.03 | Hurt |
| remove upper bound on length ratio (ablation) | length reward (Llama3.2-1B) · MATH | accuracy: 25.00 -> 8.46 | Hurt |
| remove lower bound on length ratio (ablation) | length reward (Llama3.2-1B) · GSM8K | accuracy: 52.92 -> 51.25 | Hurt |
| MetaAct-RL vs Self-Correction-RL | Llama3.2-1B · GSM8K | accuracy: 46.70 -> 52.92 | Improved |
| MetaAct-STaR instead of MetaAct-RL on small model | Llama3.2-1B · 6 benchmark average | accuracy: 42.23 -> 30.33 | Hurt |
| MetaAct-RL vs CoT-RL | Qwen2.5-1.5B · AIME-2024 | accuracy: 28.33 -> 35.83 | Improved |
| MetaAct-RL vs ADORA | Qwen2.5-1.5B · AIME-2025 | accuracy: 25.16 -> 25.42 | No clear effect |
Closest papers by meaning
- Reshaping Reasoning in LLMs: A Theoretical Analysis of RL Training Dynamics through Pattern Selection (ICLR 2026 · similarity 0.91)
- ReMA: Learning to Meta-Think for LLMs with Multi-agent Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
- On the Interplay of Pre-Training, Mid-Training, and RL on Reasoning Language Models (ICML 2026 · similarity 0.90)
- ProRL: Prolonged Reinforcement Learning Expands Reasoning Boundaries in Large Language Models (NeurIPS 2025 · similarity 0.90)
- RL Squeezes, SFT Expands: A Comparative Study of Reasoning LLMs (ICLR 2026 · similarity 0.90)
- Beyond Accuracy: Dissecting Mathematical Reasoning for LLMs Under Reinforcement Learning (NeurIPS 2025 · similarity 0.90)
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.
