arXiv：cs.LG· Shangyang Wu, Shuai Zhao, Ziyue Zhu, Jinyang Wu, Anh Tuan Luu, Haoran Luo·· 3 小时前AI 评分34
SCAD：面向长时程智能体的结构化信用分配与蒸馏
SCAD: Structured Credit Assignment and Distillation for Long-Horizon Agents
AI 导读
针对长时程智能体训练中稀疏终端奖励难以区分中间贡献、在线策略蒸馏随学生生成历史增长而丢失教师指导的问题，研究者提出 SCAD，将交互组织为规划与有界子任务执行，在局部上下文中蒸馏执行过程，并通过跨 rollout 子任务前缀树细化规划信用。在所有评测基准上，SCAD 的宏平均准确率较最强训练基线在文本任务上提升 4.48 个百分点，多模态任务上提升 4.19 个百分点。
正文
Abstract:Training long-horizon agents to solve complex tasks requires effective supervision over extended interaction sequences. However, sparse terminal rewards obscure intermediate contributions, while on-policy distillation can lose informative teacher guidance as student-generated histories grow. To address this problem, we introduce SCAD, which organizes interactions into planning and bounded subtask execution, distills execution in local contexts, and refines planning credit through cross-rollout subtask prefix trees, with planning receiving full terminal credit and execution receiving positive terminal credit and teacher guidance. Across all evaluated benchmarks, SCAD improves macro-average accuracy over the strongest training baseline by 4.48 percentage points for text tasks and 4.19 points for multimodal tasks. SCAD effectively combines outcome-based credit assignment with teacher-guided distillation to improve planning and execution in long-horizon agents.
| Comments: | 32 pages |
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.03372 [cs.LG] |
| (or arXiv:2610.03372v1 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2610.03372 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Shangyang Wu [view email]
[v1]
Fri, 2 Oct 2026 14:29:42 UTC (1,350 KB)
来源：arXiv：cs.LG · arxiv.org