{"ID":2869203,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2509.14718","arxiv_id":"2509.14718","title":"ToolSample: Dual Dynamic Sampling Methods with Curriculum Learning for RL-based Tool Learning","abstract":"While reinforcement learning (RL) is increasingly used for LLM-based tool learning, its efficiency is often hampered by an overabundance of simple samples that provide diminishing learning value as training progresses. Existing dynamic sampling techniques are ill-suited for the multi-task structure and fine-grained reward mechanisms inherent to tool learning. This paper introduces Dynamic Sampling with Curriculum Learning (DSCL), a framework specifically designed to address this challenge by targeting the unique characteristics of tool learning: its multiple interdependent sub-tasks and multi-valued reward functions. DSCL features two core components: Reward-Based Dynamic Sampling, which uses multi-dimensional reward statistics (mean and variance) to prioritize valuable data, and Task-Based Dynamic Curriculum Learning, which adaptively focuses training on less-mastered sub-tasks. Through extensive experiments, we demonstrate that DSCL significantly improves training efficiency and model performance over strong baselines, achieving a 3.29\\% improvement on the BFCLv3 benchmark. Our method provides a tailored solution that effectively leverages the complex reward signals and sub-task dynamics within tool learning to achieve superior results.","short_abstract":"While reinforcement learning (RL) is increasingly used for LLM-based tool learning, its efficiency is often hampered by an overabundance of simple samples that provide diminishing learning value as training progresses. Existing dynamic sampling techniques are ill-suited for the multi-task structure and fine-grained rew...","url_abs":"https://arxiv.org/abs/2509.14718","url_pdf":"https://arxiv.org/pdf/2509.14718v1","authors":"[\"Zihao Feng\",\"Xiaoxue Wang\",\"Bowen Wu\",\"Hailong Cao\",\"Tiejun Zhao\",\"Qun Yu\",\"Baoxun Wang\"]","published":"2025-09-18T08:04:49Z","proceeding":"cs.LG","tasks":"[\"cs.LG\",\"cs.CL\"]","methods":"[\"Reinforcement Learning\",\"Large Language Model\"]","has_code":false}