Research map / Reinforcement learning / RL from feedback
Adaptive Rollout Allocation for Online Reinforcement Learning with Verifiable Rewards
Hieu Trung Nguyen, Bao Nguyen, Wenao Ma, Yuzhi Zhao, Ruifeng She, Viet Anh Nguyen
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add VIP rollout allocation | RLOO · AIME24 (Qwen2.5-Math-1.5B, 8xQ) | Pass@32: +12.3 (18.29 -> 30.55) | Improved |
| add VIP rollout allocation | RLOO · AIME24 (Qwen2.5-Math-1.5B, 8xQ) | Mean@32: +6.3 (3.43 -> 9.68) | Improved |
| add VIP rollout allocation | Dr. GRPO · AIME24 (Qwen2.5-Math-1.5B, 8xQ) | Pass@32: 34.56 -> 33.63 | Hurt |
| add VIP rollout allocation | Dr. GRPO · AIME24 (Qwen2.5-Math-7B, 8xQ) | Pass@32: 50.0 -> 58.98 | Improved |
| add VIP rollout allocation | RLOO · AIME24 (Llama-3.2-3B-Instruct, 8xQ) | Pass@32: 28.84 -> 35.59 | Improved |
| add VIP rollout allocation | Dr. GRPO · Bamboogle (Qwen2.5-3B-Instruct) | EM: 20 -> 23.2 | Improved |
| add VIP rollout allocation | RLOO · Bamboogle (Qwen2.5-3B-Instruct) | EM: 10.4 -> 17.6 | Improved |
| add VIP rollout allocation | Dr. GRPO · MuSiQue test (Qwen2.5-3B-Instruct) | EM: 6 -> 10.5 | Improved |
| add VIP rollout allocation | RLOO · Bamboogle (Qwen2.5-3B-Instruct) | F1@5: 0.190 -> 0.264 | Improved |
| add adaptive variance-aware allocator instead of Inverse Accuracy heuristic (ablation) | RLOO+GP · AIME24 (Qwen2.5-Math-1.5B) | Pass@32: 24.61 -> 30.55 | Improved |
| add adaptive variance-aware allocator instead of Inverse Variance heuristic (ablation) | RLOO+GP · AIME24 (Qwen2.5-Math-1.5B) | Mean@32: 4.90 -> 9.68 | Improved |
| add Gaussian process predictor instead of Ridge Regression (ablation) | RLOO+VIP · AIME24 (Qwen2.5-Math-1.5B) | Pass@32: 29.90 -> 30.55 | Improved |
Closest papers by meaning
- Variance-Optimal State Resampling for Reinforcement Learning with Verifiable Rewards (NeurIPS 2026 · similarity 0.89)
- SALT: When More Rollouts Don’t Help in Group-Based Policy Optimization and How to Make Them Matter (NeurIPS 2026 · similarity 0.89)
- DUET: Optimize Token-Budget Allocation for Reinforcement Learning with Verifiable Rewards (NeurIPS 2026 · similarity 0.87)
- Act Only When It Pays: Efficient Reinforcement Learning for LLM Reasoning via Selective Rollouts (NeurIPS 2025 · similarity 0.87)
- Learn from the Gap: Differential-Aware Advantage Pruning with Adaptive Rollout Sampling for GRPO (NeurIPS 2026 · similarity 0.87)
- Leveraging Error Diversity in Group Rollouts for Reinforcement Learning (NeurIPS 2026 · similarity 0.86)
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.
