Saved in:
| Main Authors: | , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2602.14169 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917275557167104 |
|---|---|
| author | Guo, Yiran Qiao, Zhongjian Xie, Yingqi Liu, Jie Ye, Dan Zhang, Ruiqing Qiu, Shuang Xu, Lijie |
| author_facet | Guo, Yiran Qiao, Zhongjian Xie, Yingqi Liu, Jie Ye, Dan Zhang, Ruiqing Qiu, Shuang Xu, Lijie |
| contents | Effective exploration is a key challenge in reinforcement learning for large language models: discovering high-quality trajectories within a limited sampling budget from the vast natural language sequence space. Existing methods face notable limitations: GRPO samples exclusively from the root, saturating high-probability trajectories while leaving deep, error-prone states under-explored. Tree-based methods blindly disperse budgets across trivial or unrecoverable states, causing sampling dilution that fails to uncover rare correct suffixes and destabilizes local baselines. To address this, we propose Deep Dense Exploration (DDE), a strategy that focuses exploration on $\textit{pivots}$-deep, recoverable states within unsuccessful trajectories. We instantiate DDE with DEEP-GRPO, which introduces three key innovations: (1) a lightweight data-driven utility function that automatically balances recoverability and depth bias to identify pivot states; (2) local dense resampling at each pivot to increase the probability of discovering correct subsequent trajectories; and (3) a dual-stream optimization objective that decouples global policy learning from local corrective updates. Experiments on mathematical reasoning benchmarks demonstrate that our method consistently outperforms GRPO, tree-based methods, and other strong baselines. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_14169 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Deep Dense Exploration for LLM Reinforcement Learning via Pivot-Driven Resampling Guo, Yiran Qiao, Zhongjian Xie, Yingqi Liu, Jie Ye, Dan Zhang, Ruiqing Qiu, Shuang Xu, Lijie Machine Learning Artificial Intelligence Computation and Language Effective exploration is a key challenge in reinforcement learning for large language models: discovering high-quality trajectories within a limited sampling budget from the vast natural language sequence space. Existing methods face notable limitations: GRPO samples exclusively from the root, saturating high-probability trajectories while leaving deep, error-prone states under-explored. Tree-based methods blindly disperse budgets across trivial or unrecoverable states, causing sampling dilution that fails to uncover rare correct suffixes and destabilizes local baselines. To address this, we propose Deep Dense Exploration (DDE), a strategy that focuses exploration on $\textit{pivots}$-deep, recoverable states within unsuccessful trajectories. We instantiate DDE with DEEP-GRPO, which introduces three key innovations: (1) a lightweight data-driven utility function that automatically balances recoverability and depth bias to identify pivot states; (2) local dense resampling at each pivot to increase the probability of discovering correct subsequent trajectories; and (3) a dual-stream optimization objective that decouples global policy learning from local corrective updates. Experiments on mathematical reasoning benchmarks demonstrate that our method consistently outperforms GRPO, tree-based methods, and other strong baselines. |
| title | Deep Dense Exploration for LLM Reinforcement Learning via Pivot-Driven Resampling |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2602.14169 |