Retrieval-Augmented Generation (RAG) im- proves the factuality of large language models (LLMs), yet existing RAG systems often struggle with complex, multi-step reasoning that requires adaptive retrieval and continuous revision of intermediate contexts. Recent reinforcement learning (RL)-based agentic RAG methods partially alleviate this issue, but typically rely on coarse-grained action spaces and trajectory-level rewards, resulting in weak reward assignment and a bias toward short-horizon, stereotyped reasoning template. To address, we propose 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. Experiments across a diverse set of multi-hop, open-domain, and agentic reasoning benchmarks, spanning multiple backbone model sizes, demonstrate that AGENTICRAG- R1 consistently outperforms strong baselines. Moreover, AGENTICRAG-R1 learns more robust, interpretable, and memory-aware reasoning behaviors, highlighting the effect of fine-grained action modeling and information- aware optimization for long-horizon reasoning. Our code is anonymous available at https: //github.com/jiangxinke/Harness-RL/ tree/AgenticRAG-R1-Whitebox.

Rubric-based rewards offer interpretable and fine-grained optimization signals for reinforcement learning in open-ended tasks where verifiable answers are unavailable. However, pre-constructed rubrics remain static throughout training, creating a fundamental mismatch with the evolving policy: fixed criteria gradually lose discriminative power as the model improves, leading to reward saturation and potential hacking. Recent dynamic rubric methods partially address this but rely on external frontier models or ground-truth answers, and update rubrics only at coarse granularity. We propose EvoRubrics, a co-evolutionary RL framework where a Policy LLM and a Rubric Generator jointly improve through adversarial interaction within each training step. As the policy improves under the rubric generator’s guidance, the rubric generator adapts its criteria to remain discriminative and informative, enabling evaluation to track the policy in real time and naturally inducing an automatic curriculum. Experiments show that EvoRubrics consistently outperforms static and dynamic rubric baselines across benchmarks. The learned Rubric Generator further generalizes as a transferable reward model. Notably, even a fully self-supervised variant without any external supervision achieves meaningful gains, suggesting that co-evolution between generation and evaluation alone can provide sufficiently rich learning signals. Our code is publicly available at https://github.com/huangbaixiang100/EvoRubrics.

As training regimes centered on human-curated data encounter rising costs and an increasingly binding scaling ceiling, a core question in agent learning shifts from enlarging exogenous datasets to enabling agents to **self-evolve** under limited exogenous supervision. Prior surveys on self-evolution and self-improvement typically organize the literature by evolutionary stages or system architectures. While effective for summarizing implementation pipelines, these perspectives do not offer a unified explanation of what an agent actually observes during self-improvement, nor why such signals can serve as sufficient and reliable supervision. To address this gap, we introduce **endogenous observations**, defined as learning signals that arise from an agent’s own generation, evaluation, and interaction processes rather than exogenous labels. We systematize existing approaches into three categories—model-internal, peer-induced, and scenario-grounded observations—according to how these signals are produced, and we summarize the characteristic properties and typical application settings of each category. Finally, we discuss key challenges in reliability, scalability, and long-horizon stability, and outline open directions toward robust and sustainable self-evolving agents.

Open-ended deep research (OEDR) requires systems to acquire knowledge through multi-round retrieval and generate coherent long-form reports. The outline plays a central role as a structural scaffold that coordinates retrieval, evidence organization, and generation. However, existing methods either fix the outline before writing or refine it with local heuristics, leading to scaffold drift under continuous information accumulation and delayed feedback for evaluating outline modifications. We propose ScaffoldAgent, a utility-guided dynamic outline optimization framework for OEDR. ScaffoldAgent models outline evolution as a structured decision process with three operations: Expansion, Contraction, and Revision, enabling controlled updates to the report scaffold. It further introduces a utility-guided feedback mechanism that estimates the downstream value of each outline operation from retrieval gain, structural coherence, and trial-generation quality. The resulting utility signal guides node selection, operation scheduling, and termination during inference. Experiments on DeepResearch Bench and DeepResearch Gym show that ScaffoldAgent consistently improves long-form report generation and factual grounding over existing deep research agents.

Process reward models (PRMs) enhance the reasoning capabilities of large language models (LLMs) by providing fine-grained feedback, yet training PRMs typically requires expensive stepwise annotations. Outcome-supervised PRMs offer a scalable alternative by learning from final-answer correctness alone, but this introduces a fundamental *credit assignment* challenge, i.e., attributing outcomes to responsible reasoning steps. Existing approaches rely on either uniform or causal assignment, both of which fail to anchor credit in step correctness and thus hinder process error identification.

In this work, we propose Outcome-Supervised Process Reward Modeling via **L**earnable **C**redit **A**ssignment (**LCA**), an outcome-supervised PRM framework that jointly learns credit assignment and reward modeling under the principle of *Weakest Link Assignment: a reasoning chain is as strong as its weakest link*. To address mutual dependence between credit assignment and reward modeling, we formalize outcome-supervised PRM as a Multiple Instance Learning (MIL) problem and introduce Softmax-Weighted-Sum (SWS) pooling, an MIL pooling technique tailored for strong dependence and redundancy among reasoning states. We establish instance-level Bayes consistency of the resulting learning objective under mild assumptions. Extensive experiments demonstrate that **LCA** outperforms state-of-the-art outcome-supervised PRMs across multiple tasks and backbones. Code is available at https://github.com/JiaTianyu20031016/MIL.git.

Schema linking is a crucial step in Text-to-SQL systems, connecting natural language mentions to relevant database tables and columns. Despite recent advances in LLM-based Text-to-SQL models, current approaches face two fundamental limitations: (1) providing detailed column descriptions inflates the input context, often exceeding the model’s effective sequence length, and (2) predicting multi-token column names increases output complexity, making the task fragile to generation errors. To address these challenges, we propose SchemaToken, which represents each database column as a dedicated token in the LLM’s vocabulary. This design compresses column semantics into single embeddings, reducing context length and simplifying prediction. We introduce a two-stage training paradigm: (i) Semantic Alignment, which enables the LLM to understand the newly added schema tokens via structured embeddings, and (ii) Pseudo-task Alignment, which trains the LLM to generate these tokens as an unordered set using reinforcement learning with a Jaccard-based reward. Experiments on multiple benchmark datasets demonstrate that SchemaToken significantly improves schema linking performance, particularly for large and complex schemas, while maintaining computational efficiency. Our code is publicly available at https://github.com/MrBlankness/SchemaToken.

Interactive medical questioning is essential in clinical consultations, where physicians must actively gather necessary patient information. Yet existing medical Large Language Models (LLMs) predominantly follow a reactive paradigm, risking diagnostic errors by answering before seeking sufficient details. To bridge this gap, we propose ProMed, a reinforcement learning framework that transitions LLMs toward a proactive paradigm, enabling them to ask clinically valuable questions before decision-making. Central to ProMed is the Shapley Information Gain (SIG) reward, which quantifies a question’s clinical utility as the amount of newly acquired information, while considering its contextual importance via Shapley values. We integrate SIG into a two-stage training pipeline: (1) SIG-Guided Model Initialization uses Monte Carlo Tree Search to construct high-reward interaction trajectories for supervision, and (2) SIG-Augmented Policy Optimization, with a novel SIG-guided Reward Distribution Mechanism that prioritizes informative questions for fine-grained optimization. Experiments on partial-information medical benchmarks show that ProMed significantly outperforms state-of-the-art methods by 6.29% on average and delivers a 54.45% gain over the reactive paradigm, and generalizes robustly to out-of-domain cases. Our codes are available at https://github.com/hxxding/ProMed.

Federated Retrieval (FR) routes queries across multiple external knowledge sources, to mitigate hallucinations of LLMs, when necessary external knowledge is distributed. However, existing methods struggle to retrieve high-quality and relevant documents for ambiguous queries, especially in cross-domain scenarios, which significantly limits their effectiveness in supporting downstream generation tasks. Inspired by Dynamic Information Flow (DIF), we propose DFAMS, a novel framework that leverages DIF to identify latent query intents and construct semantically aligned knowledge partitions for accurate retrieval across heterogeneous sources. Specifically, DFAMS probes the DIF in LLMs by leveraging gradient signals from a few annotated queries and employing Shapley value-based attribution to trace neuron activation paths associated with intent recognition and subdomain boundary detection. Then, DFAMS leverages DIF to train an alignment module via multi-prototype contrastive learning, enabling fine-grained intra-source modeling and inter-source semantic alignment across knowledge bases. Experimental results across five benchmarks show that DFAMS outperforms advanced FR methods by up to 14.37% in knowledge classification accuracy, 5.38% in retrieval recall, and 6.45% in downstream QA accuracy, demonstrating its effectiveness in complex FR scenarios. Our code is publicly available at https://github.com/Artessay/DFAMS.

Large Language Models (LLMs) excel in general language tasks, motivating their adaptation to specialized domains such as healthcare. Effective domain adaptation typically involves supervised fine-tuning (SFT) on carefully selected instruction-tuning data. Current data selection methods adopt a data-centric approach, relying on external annotations and heuristics to identify externally defined high-quality or challenging data. Our exploratory experiments highlight this approach fails to improve the model’s domain performance, due to misalignment between selected data and the model’s knowledge distribution. To tackle this, we propose Decomposed Difficulty-based Data Selection (3DS), a two-stage model-centric data selection framework that aligns data selection with the model’s distribution. 3DS employs Prompt-Driven Data Selection to filter out noise based on the model’s knowledge via explicit alignment in Stage#1, then adopts Decomposed Difficulty-based Data Selection to guide selection via three novel data difficulty metrics, including Instruction Understanding, Response Confidence, and Response Correctness in Stage#2, enhanced by an attention-based importance weighting mechanism for accurate calibration.Extensive experiments in the healthcare domain show 3DS outperforms existing methods by up to 2.97% accuracy, with additional validation in law and general domains, confirming its generalization ability. Our dataset and code are open-sourced at https://github.com/PuppyKnightUniversity/3DS.

In this paper, we investigate the retrieval-augmented generation (RAG) based on Knowledge Graphs (KGs) to improve the accuracy and reliability of Large Language Models (LLMs). Recent approaches suffer from insufficient and repetitive knowledge retrieval, tedious and time-consuming query parsing, and monotonous knowledge utilization. To this end, we develop a Hypothesis Knowledge Graph Enhanced (HyKGE) framework, which leverages LLMs’ powerful reasoning capacity to compensate for the incompleteness of user queries, optimizes the interaction process with LLMs, and provides diverse retrieved knowledge. Specifically, HyKGE explores the zero-shot capability and the rich knowledge of LLMs with Hypothesis Outputs to extend feasible exploration directions in the KGs, as well as the carefully curated prompt to enhance the density and efficiency of LLMs’ responses. Furthermore, we introduce the HO Fragment Granularity-aware Rerank Module to filter out noise while ensuring the balance between diversity and relevance in retrieved knowledge. Experiments on two Chinese medical multiple-choice question datasets and one Chinese open-domain medical Q&A dataset with two LLM turbos demonstrate the superiority of HyKGE in terms of accuracy and explainability. Code is available at https://github.com/Artessay/HyKGE.

Retrieval-Augmented Generation (RAG) offers an effective solution to the issues faced by Large Language Models (LLMs) in hallucination generation and knowledge obsolescence by incorporating externally retrieved knowledge. However, existing methods lack effective control mechanisms for integrating internal and external knowledge. Inspired by human cognitive processes, we propose Parenting, a novel framework that decouples, identifies, and purposefully optimizes parameter subspaces related to adherence and robustness. Specifically, Parenting utilizes a key parameter mining method that combines forward and backward propagation signals to localize subspaces representing different capabilities. Then, Parenting employs a type-tailored tuning strategy, applying specific and appropriate optimizations to different subspaces, aiming to achieve a balanced enhancement of both adherence and robustness. Extensive experiments on various datasets and models validate the effectiveness and generalizability of our method. Our code is available at https://github.com/Nostradamus4869/Parenting.

In the pursuit of enhancing domain-specific Large Language Models (LLMs), Retrieval-Augmented Generation (RAG) emerges as a promising solution to mitigate issues such as hallucinations, outdated knowledge, and limited expertise in highly specialized queries. However, existing approaches to RAG fall short by neglecting system state variables, which are crucial for ensuring adaptive control, retrieval halting, and system convergence. In this paper, we introduce the Turing-Complete-RAG (TC-RAG) through rigorous proof, a novel framework that addresses these challenges by incorporating a Turing Complete System to manage state variables, thereby enabling more efficient and accurate knowledge retrieval. By leveraging a memory stack system with adaptive retrieval, reasoning, and planning capabilities, TC-RAG not only ensures the controlled halting of retrieval processes but also mitigates the accumulation of erroneous knowledge via Push and Pop actions. In the case study of the medical and general domain, our extensive experiments on seven real-world healthcare and general-domain datasets demonstrate the superiority of TC-RAG over existing methods in accuracy by over 7.20%. Our code, datasets and RAG resources have been available at https://github.com/Artessay/TC-RAG.

Extracting structured knowledge from unstructured text data has a wide range of application prospects, and a pervasive trend is to develop text annotation tools to help extraction. However, they often encounter issues such as single scenario usage, lack of effective human-machine collaboration, insufficient model supervision, and suboptimal utilization of Large Language Models (LLMs). We introduces an interactive unstructured text annotation and knowledge extraction system that synergistically integrates LLMs and ModelOps to alleviate these issues. The system leverages LLMs for enhanced performance in low-resource contexts, employs a ModelOps platform to monitor models throughout their lifecycle, and amalgamates interactive annotation methods with online machine learning and active learning. The demo video and website are now publicly available.

Efforts have been made to apply topic seed words to improve the topic interpretability of topic models. However, due to the semantic diversity of natural language, supervisions from seed words could be ambiguous, making it hard to be incorporated into the current neural topic models. In this paper, we propose SeededNTM, a neural topic model enhanced with supervisions from seed words on both word and document levels. We introduce a context-dependency assumption to alleviate the ambiguities with context document information, and an auto-adaptation mechanism to automatically balance between multi-level information. Moreover, an intra-sample consistency regularizer is proposed to deal with noisy supervisions via encouraging perturbation and semantic consistency. Extensive experiments on multiple datasets show that SeededNTM can derive semantically meaningful topics and outperforms the state-of-the-art seeded topic models in terms of topic quality and classification accuracy.