Large language model agents increasingly rely on long-horizon reasoning to solve complex tasks involving planning, tool use, and memory. A critical capability in such settings is reflection: assessing trajectory progress, identifying missing evidence and unreliable intermediate states, and deciding whether to continue, revise, or abandon the current branch. Learning effective reflection, however, is challenging because reflection is performed locally within the current branch, whereas its utility can only be determined by its contribution to the final trajectory outcome. This local-global mismatch makes outcome-based reinforcement learning provide only local, sparse and delayed supervision for reflective decisions. To solve these, we propose LoongReflect, a training framework that formulates reflection as a memory-control policy. The agent operates over a reversible trajectory tree using explicit reflect and backtrack actions. Reflection consolidates verified facts, missing evidence, and branch-specific risks into working memory, while backtracking removes an unreliable branch from the active context and preserves a concise corrective lesson. To learn this policy, LoongReflect combines two complementary signals through a look-ahead, extragradient-style coordination mechanism. A fast channel distills globally informed reflective behavior from a privileged teacher, with supervision restricted to reflection and backtracking tokens. A slow channel optimizes complete trajectories using outcome-based GRPO, aligning local control decisions with final task success. Experiments on multi-hop retrieval-augmented generation and mathematical reasoning benchmarks demonstrate consistent improvements over outcome-only reinforcement learning and self-distillation baselines.
Zhixin Zhang, Xinke Jiang, Zhibang Yang et al.· 0 citations
With the popularization of Electronic Health Records (EHR), the emergency system has stored a large number of historical dispatch records, which can provide valuable insights for the optimization of current pre-hospital care. However, the inconvenient interaction manner of cur-rent information retrieval systems hinders researchers from exploring these historical records. To address this issue, we propose a novel framework that leverages the language understanding and code generation ability of Large Language Models (LLMs) to build an information retrieval system with Visualization-oriented Natural-language-based Inter-faces (V-NLI). To incorporate both domain-specific and task-related prior knowledge, we generate the instruction datasets based on the ability of closed-source LLMs in a multi-stage manner and conduct supervised fine-tuning on open-source LLMs. We also devised various mechanisms for augmenting the capabilities of open-source LLMs in query interpretation and code generation. To validate the effectiveness and generalizability of our framework, we conducted experiments on a public dataset NLV. More significantly, we performed more detailed experiments on a dataset including over 1 mil-lion pre-hospital emergency historical records in ten years. The performance of our method surpasses all baseline methods and achieves comparable results even with some SOTA closed-source models.
Xin Gao, Zheng-Ye Zhu, Xin-Yu Ma et al.· Tsinghua Science and Technol...· 0 citations
Harness-RL is introduced, a structured reinforcement learning framework that combines Conflict-Aware Policy Optimization (CAPO) with interface-level black-box trajectory construction and supports both central-only and joint multi-agent training.
Xinke Jiang, Zhixin Zhang, Zhibang Yang et al.· 0 citations
ScaffoldAgent is proposed, a utility-guided dynamic outline optimization framework for OEDR that models outline evolution as a structured decision process with three operations: Expansion, Contraction, and Revision, enabling controlled updates to the report scaffold.
This work proposes AgenticRag-R1, a RL framework that deeply integrates reasoning, retrieval, and memory via a memory stack and fine-grained action space, supported by hierarchical action-aware rewards and an information-aware trajectory rejection strategy to enable effective long-horizon learning.
Xinke Jiang, Yue Fang, Zhibang Yang et al.· 1 citation
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
