AGENTIC R AG-R1: Agentic Reinforcement Learning with Stack Memory for Multi-Step Reasoning, Retrieval and Memorizing · CiteArk

AGENTIC R AG-R1: Agentic Reinforcement Learning with Stack Memory for Multi-Step Reasoning, Retrieval and Memorizing

公开

作者：Xinke Jiang、Yue Fang、Zhibang Yang、Jiaran Gao、Zhixin Zhang、Tao Feng、Rihong Qiu、Wentao Zhang、Hongxin Ding、Ruizhe Zhang、Yongxin Xu、Yuheng Huang、Xu Chu、Junfeng Zhao、Yasha Wang

arXiv 2026 人工智能机器学习 (cs.LG)多智能体系统机器学习 (stat.ML)

复现结果

全部结论 ↗

尚无复现结果

论文中的结论可在「研究结论」中查看。

完整研究主张目录

0 / 4 条结论已通过验证

其余仍在验证中

关键指标论文报告值 → 复现观测值 · 点击行展开详情

AGENTIC R AG-R1 reports higher or near-highest F1 than listed baselines across seven multi-hop, open-domain, and agentic QA benchmarks.整组结论：实验受阻，详见具体原因0/3 项指标获得支持 · 0 项已有评估3项测量结果展开结果表

No verified repository, trained checkpoint, complete curated training data, retrieval implementation, or strict evaluator is available in the fixed inputs.

测量结果

已评估 0/3 项

获得支持 0有争议或冲突 0尚未评估 3

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
datasets: ["2WikiMultiHopQA","HotpotQA","Bamboogle","FRAMES","MusiQue","NQ","TriviaQA"] · aggregation: arithmetic average · model_variant: Qwen2.5-3B	33.55 percentage_points	—	尚未评估
split: test · dataset: HotpotQA · aggregation: single reported evaluation · model_variant: Qwen2.5-3B	44 percentage_points	—	尚未评估
split: test · dataset: Bamboogle · aggregation: single reported evaluation · model_variant: Qwen2.5-7B	49.21 percentage_points	—	尚未评估

打开 Claim 详情

The paper reports increasing AGENTIC R AG-R1 F1 as the maximum reasoning-step budget grows on selected 2Wiki, Bamboogle, and TriviaQA evaluations.整组结论：实验受阻，详见具体原因0/2 项指标获得支持 · 0 项已有评估2项测量结果展开结果表

The missing checkpoint, retrieval service, and complete evaluation protocol prevent strict comparable evidence; execution resources are not the blocker.

测量结果

已评估 0/2 项

获得支持 0有争议或冲突 0尚未评估 2

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
dataset: 2WikiMultiHopQA · max_steps: 10 · aggregation: single reported evaluation · model_variant: Qwen2.5-3B	23.91 percentage_points	—	尚未评估
dataset: 2WikiMultiHopQA · max_steps: 30 · aggregation: single reported evaluation · model_variant: Qwen2.5-3B	47.45 percentage_points	—	尚未评估

打开 Claim 详情

The paper reports lower average inference time than TC-RAG and downstream results on MedQA, DeepResearch Bench, ALFWorld, and WebShop.整组结论：实验受阻，详见具体原因0/3 项指标获得支持 · 0 项已有评估3项测量结果展开结果表

The original hardware-sensitive benchmark conditions are underspecified by the fixed paper and repository, so a strict comparable verdict is not available.

测量结果

已评估 0/3 项

获得支持 0有争议或冲突 0尚未评估 3

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
hardware: NVIDIA A800 server · aggregation: average · model_variant: Qwen2.5-3B	10.15 seconds	—	尚未评估
hardware: NVIDIA A800 server · aggregation: average · model_variant: Qwen2.5-3B	11.86 seconds	—	尚未评估
judge: Qwen2.5-72B · dataset: MedQA test · model_variant: Qwen2.5-7B-Instruct	87 percentage_points	—	尚未评估

打开 Claim 详情

The paper reports that removing memory actions, rollout rejection, or both lowers average F1, and that removing both retrieval and memory rewards gives the lowest ablation average.整组结论：实验受阻，详见具体原因0/2 项指标获得支持 · 0 项已有评估2项测量结果展开结果表

The trained variants, reward model prompts, curated training set, optimizer implementation, and exact seeds are unavailable.

测量结果

已评估 0/2 项

获得支持 0有争议或冲突 0尚未评估 2

保留同一结论下的完整结果组。已有评估排在前面；单项判断不替代整组结论。展开行查看条件、来源与历次证据。

每行是一项测量，分别显示论文报告、最近可用实测和累计证据状态。
指标与条件	论文报告	最近可用实测	证据判断
condition: full model · aggregation: average across seven benchmarks · model_variant: Qwen2.5-3B	33.55 percentage_points	—	尚未评估
condition: without retrieval and memory rewards · aggregation: average across seven benchmarks · model_variant: Qwen2.5-3B	19.94 percentage_points	—	尚未评估

打开 Claim 详情