Learn the Ropes, Then Trust the Wins: Self-imitation with Progressive Exploration for Agentic Reinforcement Learning

Reinforcement learning (RL) is the dominant paradigm for sharpening strategic tool use capabilities of LLMs on long-horizon, sparsely-rewarded agent tasks, yet it faces a fundamental challenge of exploration-exploitation trade-off. Existing studies stimulate exploration through the lens of policy entropy, but such mechanical entropy maximization is prone to RL training instability due to the multi-turn distribution shifting. In this paper, we target the progressive exploration-exploitation balance under the guidance of the agent own experiences without succumbing to either entropy collapsing or runaway divergence. We propose SPEAR, a curriculum-based self-imitation learning (SIL) recipe for training agentic LLMs. It extends the vanilla SIL framework, where a replay buffer stores self-generated promising trajectories for off-policy update, by gradually steering the policy evolution within a well-balanced range of entropy across stages. Specifically, our approach incorporates a curriculum to manage the exploration process, utilizing intrinsic rewards to foster skill-level exploration and facilitating action-level exploration through SIL. At first, the auxiliary tool call reward plays a critical role in the accumulation of tool-use skills, enabling broad exposure to the unfamiliar distributions of the environment feedback with an upward entropy trend. As training progresses, self-imitation gets strengthened to exploit existing successful patterns from replayed experiences for comparative action-level exploration, accelerating solution iteration without unbounded entropy growth. To further stabilize training, we recalibrate the advantages of experiences in the replay buffer to address the potential policy drift. Reugularizations such as the clipping of tokens with high covariance between probability and advantage are introduced to the trajectory-level entropy control to curb over-confidence.

The Choice of Divergence: A Neglected Key to Mitigating Diversity Collapse in Reinforcement Learning with Verifiable Reward

A central paradox in fine-tuning Large Language Models (LLMs) with Reinforcement Learning with Verifiable Reward (RLVR) is the frequent degradation of multi-attempt performance (Pass@k) despite improvements in single-attempt accuracy (Pass@1). This is often accompanied by catastrophic forgetting, where models lose previously acquired skills. While various methods have been proposed, the choice and function of the divergence term have been surprisingly unexamined as a proactive solution. We argue that standard RLVR objectives -- both those using the mode-seeking reverse KL-divergence and those forgoing a divergence term entirely -- lack a crucial mechanism for knowledge retention. The reverse-KL actively accelerates this decay by narrowing the policy, while its absence provides no safeguard against the model drifting from its diverse knowledge base. We propose a fundamental shift in perspective: using the divergence term itself as the solution. Our framework, Diversity-Preserving Hybrid RL (DPH-RL), leverages mass-covering f-divergences (like forward-KL and JS-divergence) to function as a rehearsal mechanism. By continuously referencing the initial policy, this approach forces the model to maintain broad solution coverage. Extensive experiments on math and SQL generation demonstrate that DPH-RL not only resolves the Pass@k degradation but improves both Pass@1 and Pass@k in- and out-of-domain. Additionally, DPH-RL is more training-efficient because it computes f-divergence using generator functions, requiring only sampling from the initial policy and no online reference model. Our work highlights a crucial, overlooked axis for improving RLVR, demonstrating that the proper selection of a divergence measure is a powerful tool for building more general and diverse reasoning models.

Guess What I am Thinking: A Benchmark for Inner Thought Reasoning of Role-Playing Language Agents

Recent advances in LLM-based role-playing language agents (RPLAs) have attracted broad attention in various applications. While chain-of-thought reasoning has shown importance in many tasks for LLMs, the internal thinking processes of RPLAs remain unexplored. Understanding characters' inner thoughts is crucial for developing advanced RPLAs. In this paper, we introduce ROLETHINK, a novel benchmark constructed from literature for evaluating character thought generation. We propose the task of inner thought reasoning, which includes two sets: the gold set that compares generated thoughts with original character monologues, and the silver set that uses expert synthesized character analyses as references. To address this challenge, we propose MIRROR, a chain-of-thought approach that generates character thoughts by retrieving memories, predicting character reactions, and synthesizing motivations. Through extensive experiments, we demonstrate the importance of inner thought reasoning for RPLAs, and MIRROR consistently outperforms existing methods. Resources are available at https://github.com/airaer1998/RPA_Thought.

AURORA:Automated Training Framework of Universal Process Reward Models via Ensemble Prompting and Reverse Verification

The reasoning capabilities of advanced large language models (LLMs) like o1 have revolutionized artificial intelligence applications. Nevertheless, evaluating and optimizing complex reasoning processes remain significant challenges due to diverse policy distributions and the inherent limitations of human effort and accuracy. In this paper, we present AURORA, a novel automated framework for training universal process reward models (PRMs) using ensemble prompting and reverse verification. The framework employs a two-phase approach: First, it uses diverse prompting strategies and ensemble methods to perform automated annotation and evaluation of processes, ensuring robust assessments for reward learning. Second, it leverages practical reference answers for reverse verification, enhancing the model's ability to validate outputs and improving training accuracy. To assess the framework's performance, we extend beyond the existing ProcessBench benchmark by introducing UniversalBench, which evaluates reward predictions across full trajectories under diverse policy distribtion with long Chain-of-Thought (CoT) outputs. Experimental results demonstrate that AURORA enhances process evaluation accuracy, improves PRMs' accuracy for diverse policy distributions and long-CoT responses. The project will be open-sourced at https://auroraprm.github.io/. The Universal-PRM-7B is available at https://huggingface.co/infly/Universal-PRM-7B.

Refine Knowledge of Large Language Models via Adaptive Contrastive Learning

How to alleviate the hallucinations of Large Language Models (LLMs) has always been the fundamental goal pursued by the LLMs research community. Looking through numerous hallucination-related studies, a mainstream category of methods is to reduce hallucinations by optimizing the knowledge representation of LLMs to change their output. Considering that the core focus of these works is the knowledge acquired by models, and knowledge has long been a central theme in human societal progress, we believe that the process of models refining knowledge can greatly benefit from the way humans learn. In our work, by imitating the human learning process, we design an Adaptive Contrastive Learning strategy. Our method flexibly constructs different positive and negative samples for contrastive learning based on LLMs' actual mastery of knowledge. This strategy helps LLMs consolidate the correct knowledge they already possess, deepen their understanding of the correct knowledge they have encountered but not fully grasped, forget the incorrect knowledge they previously learned, and honestly acknowledge the knowledge they lack. Extensive experiments and detailed analyses on widely used datasets demonstrate the effectiveness of our method.

SCP-116K: A High-Quality Problem-Solution Dataset and a Generalized Pipeline for Automated Extraction in the Higher Education Science Domain

Recent breakthroughs in large language models (LLMs) exemplified by the impressive mathematical and scientific reasoning capabilities of the o1 model have spotlighted the critical importance of high-quality training data in advancing LLM performance across STEM disciplines. While the mathematics community has benefited from a growing body of curated datasets, the scientific domain at the higher education level has long suffered from a scarcity of comparable resources. To address this gap, we present SCP-116K, a new large-scale dataset of 116,756 high-quality problem-solution pairs, automatically extracted from heterogeneous sources using a streamlined and highly generalizable pipeline. Our approach involves stringent filtering to ensure the scientific rigor and educational level of the extracted materials, while maintaining adaptability for future expansions or domain transfers. By openly releasing both the dataset and the extraction pipeline, we seek to foster research on scientific reasoning, enable comprehensive performance evaluations of new LLMs, and lower the barrier to replicating the successes of advanced models like o1 in the broader science community. We believe SCP-116K will serve as a critical resource, catalyzing progress in high-level scientific reasoning tasks and promoting further innovations in LLM development. The dataset and code are publicly available at https://github.com/AQA6666/SCP-116K-open.

An Attentive Dual-Encoder Framework Leveraging Multimodal Visual and Semantic Information for Automatic OSAHS Diagnosis

Obstructive sleep apnea-hypopnea syndrome (OSAHS) is a common sleep disorder caused by upper airway blockage, leading to oxygen deprivation and disrupted sleep. Traditional diagnosis using polysomnography (PSG) is expensive, time-consuming, and uncomfortable. Existing deep learning methods using facial image analysis lack accuracy due to poor facial feature capture and limited sample sizes. To address this, we propose a multimodal dual encoder model that integrates visual and language inputs for automated OSAHS diagnosis. The model balances data using randomOverSampler, extracts key facial features with attention grids, and converts physiological data into meaningful text. Cross-attention combines image and text data for better feature extraction, and ordered regression loss ensures stable learning. Our approach improves diagnostic efficiency and accuracy, achieving 91.3% top-1 accuracy in a four-class severity classification task, demonstrating state-of-the-art performance. Code will be released upon acceptance.

LMGT: Optimizing Exploration-Exploitation Balance in Reinforcement Learning through Language Model Guided Trade-offs

The uncertainty inherent in the environmental transition model of Reinforcement Learning (RL) necessitates a careful balance between exploration and exploitation to optimize the use of computational resources for accurately estimating an agent's expected reward. Achieving balance in control systems is particularly challenging in scenarios with sparse rewards. However, given the extensive prior knowledge available for many environments, it is redundant to begin learning from scratch in such settings. To address this, we introduce \textbf{L}anguage \textbf{M}odel \textbf{G}uided \textbf{T}rade-offs (i.e., \textbf{LMGT}), a novel, sample-efficient framework that leverages the comprehensive prior knowledge embedded in Large Language Models (LLMs) and their adeptness at processing non-standard data forms, such as wiki tutorials. LMGT proficiently manages the exploration-exploitation trade-off by employing reward shifts guided by LLMs, which direct agents' exploration endeavors, thereby improving sample efficiency. We have thoroughly tested LMGT across various RL tasks and deployed it in industrial-grade RL recommendation systems, where it consistently outperforms baseline methods. The results indicate that our framework can significantly reduce the time cost required during the training phase in RL.

CogniDual Framework: Self-Training Large Language Models within a Dual-System Theoretical Framework for Improving Cognitive Tasks

Cognitive psychology investigates perception, attention, memory, language, problem-solving, decision-making, and reasoning. Kahneman's dual-system theory elucidates the human decision-making process, distinguishing between the rapid, intuitive System 1 and the deliberative, rational System 2. Recent advancements have positioned large language Models (LLMs) as formidable tools nearing human-level proficiency in various cognitive tasks. Nonetheless, the presence of a dual-system framework analogous to human cognition in LLMs remains unexplored. This study introduces the \textbf{CogniDual Framework for LLMs} (CFLLMs), designed to assess whether LLMs can, through self-training, evolve from deliberate deduction to intuitive responses, thereby emulating the human process of acquiring and mastering new information. Our findings reveal the cognitive mechanisms behind LLMs' response generation, enhancing our understanding of their capabilities in cognitive psychology. Practically, self-trained models can provide faster responses to certain queries, reducing computational demands during inference.

Promoting Equality in Large Language Models: Identifying and Mitigating the Implicit Bias based on Bayesian Theory

Large language models (LLMs) are trained on extensive text corpora, which inevitably include biased information. Although techniques such as Affective Alignment can mitigate some negative impacts of these biases, existing prompt-based attack methods can still extract these biases from the model's weights. Moreover, these biases frequently appear subtly when LLMs are prompted to perform identical tasks across different demographic groups, thereby camouflaging their presence. To address this issue, we have formally defined the implicit bias problem and developed an innovative framework for bias removal based on Bayesian theory, Bayesian-Theory based Bias Removal (BTBR). BTBR employs likelihood ratio screening to pinpoint data entries within publicly accessible biased datasets that represent biases inadvertently incorporated during the LLM training phase. It then automatically constructs relevant knowledge triples and expunges bias information from LLMs using model editing techniques. Through extensive experimentation, we have confirmed the presence of the implicit bias problem in LLMs and demonstrated the effectiveness of our BTBR approach.