Forecasting future links is a central task in temporal graph (TG) reasoning, requiring models to leverage historical interactions to predict upcoming ones. Traditional neural approaches, such as temporal graph neural networks, achieve strong performance but lack explainability and cannot be applied to unseen graphs without retraining. Recent studies have begun to explore using large language models (LLMs) for graph reasoning, but most of them are constrained to static graphs or small synthetic TGs and lack the evaluation of the quality of reasoning traces generated by LLMs. In this work, we present Reasoning-Enhanced Learning for Temporal Graphs (ReaL-TG), a reinforcement learning framework that fine-tunes LLMs to perform explainable link forecasting on real-world TGs. ReaL-TG uses outcome-based reward to encourage models to self-explore reasoning strategies from graph structure and to produce explanations that directly justify their predictions. To enable evaluation on LLM-generated reasoning traces, we propose a new evaluation protocol combining ranking metrics with an LLM-as-a-Judge system that assesses both the quality of reasoning and the impact of hallucinations. Experiments with ReaL-TG-4B, obtained by fine-tuning Qwen3-4B under our framework, show that it outperforms much larger frontier LLMs, including GPT-5 mini, on ranking metrics, while producing high-quality explanations confirmed by both the LLM judge and human evaluation.

Reinforcement Learning has emerged as a dominant post-training approach to elicit agentic RAG behaviors such as search and planning from language models. Despite its success with larger models, applying RL to compact models (e.g., 0.5--1B parameters) presents unique challenges. The compact models exhibit poor initial performance, resulting in sparse rewards and unstable training. To overcome these difficulties, we propose Distillation-Guided Policy Optimization (DGPO), which employs cold-start initialization from teacher demonstrations and continuous teacher guidance during policy optimization. To understand how compact models preserve agentic behavior, we introduce Agentic RAG Capabilities (ARC), a fine-grained metric analyzing reasoning, search coordination, and response synthesis. Comprehensive experiments demonstrate that DGPO enables compact models to achieve sophisticated agentic search behaviors, even outperforming the larger teacher model in some cases. DGPO makes agentic RAG feasible in computing resource-constrained environments.

Long-context question answering over narrative tasks is challenging because correct answers often hinge on reconstructing a coherent timeline of events while preserving contextual f low in a limited context window. Retrievalaugmented generation (RAG) methods aim to address this challenge by selectively retrieving only necessary document segments. However, narrative texts possess unique characteristics that limit the effectiveness of these existing approaches. Specifically, understanding narrative texts requires more than isolated segments, as the broader context and sequential relationships between segments are crucial for comprehension. To address these limitations, we propose ChronoRAG, a novel RAG framework specialized for narrative texts. This approach focuses on two essential aspects: refining dispersed document information into coherent and structured passages and preserving narrative flow by explicitly capturing and maintaining the temporal order among retrieved passages. We empirically demonstrate the effectiveness of ChronoRAG through experiments on the NarrativeQA and GutenQAdataset, showing substantial improvements in tasks requiring both factual identification and comprehension of complex sequential relationships, underscoring that reasoning over temporal order is crucial in resolving narrative QA.

Predicting a patient's length of stay (LOS) in the intensive care unit (ICU) is a critical task for hospital resource management, yet remains challenging due to the heterogeneous and irregularly sampled nature of electronic health records (EHRs). In this work, we propose S$^2$G-Net, a novel neural architecture that unifies state-space sequence modeling with multi-view Graph Neural Networks (GNNs) for ICU LOS prediction. The temporal path employs Mamba state-space models (SSMs) to capture patient trajectories, while the graph path leverages an optimized GraphGPS backbone, designed to integrate heterogeneous patient similarity graphs derived from diagnostic, administrative, and semantic features. Experiments on the large-scale MIMIC-IV cohort dataset show that S$^2$G-Net consistently outperforms sequence models (BiLSTM, Mamba, Transformer), graph models (classic GNNs, GraphGPS), and hybrid approaches across all primary metrics. Extensive ablation studies and interpretability analyses highlight the complementary contributions of each component of our architecture and underscore the importance of principled graph construction. These results demonstrate that S$^2$G-Net provides an effective and scalable solution for ICU LOS prediction with multi-modal clinical data. The code can be found at https://github.com/ShuqiZi1/S2G-Net.

Climate change is amplifying extreme weather and climate events worldwide [1]. Anthropogenic greenhouse gas emissions have disrupted the Earth's climate system, driving more frequent and severe heatwaves [2], cold spells [3], heavy precipitation [4], agricultural droughts [5], and tropical cyclones (TCs) [6]. Between 2016 and 2024, daily land temperature records show that extreme heat events occurred over four times more often than expected, while cold records declined by half [7]. These unprecedented shifts threaten human health [8, 9], infrastructure [10, 11], food security [12], biodiversity [13], and global economies [14, 15]. Therefore, reliable climate projections are essential for effective mitigation and adaptation strategies [16-18]. The Coupled Model Intercomparison Project (CMIP) [19] provides a foundation for global climate projections. Since its launch in 1995, CMIP has coordinated systematic evaluation of coupled general circulation models (GCMs). CMIP5 introduced Representative Concentration Pathways (RCPs), while CMIP6 extended this framework by incorporating Shared Socioeconomic Pathways (SSPs) through ScenarioMIP, enabling consistent simulations of emissions and socioeconomic trajectories to 2100 and facilitating integrated assessment of climate risks [20]. These advances have greatly enhanced the scientific and policy relevance of climate projections.

Large Language Models (LLMs) are evolving from text generators into reasoning agents. This transition makes their ability to use external tools a critical capability. However, evaluating this skill presents a significant challenge. Existing benchmarks are often limited by their reliance on synthetic tools and severely constrained action spaces. To address these limitations, we introduce MCPV erse, an expansive, real-world benchmark for evaluating agentic tool use. MCPV erse integrates more than 550 real-world, executable tools to create an unprecedented action space exceeding 147k tokens, and employs outcome-based evaluation with real-time ground truth for time-sensitive tasks. We benchmarked the state-of-the-art LLMs across three modes (Oracle, Standard, and Max-Scale), revealing that while most models suffer performance degradation when confronted with larger tool sets, the agentic models, such as Claude-4-Sonnet, can effectively leverage expanded tool spaces to improve accuracy. This finding not only exposes the limitations of state-of-the-art models in complex, real-world scenarios but also establishes MCPV erse as a critical benchmark for measuring and advancing agentic tool use capabilities. The ability of Large Language Models (LLMs) to interact with external tools, typically through function calling, is fundamental to their application in real-world scenarios. This capability allows them to access live data, execute code, and operate other systems, thereby moving beyond their static knowledge. Despite its importance, the evaluation of tool use is hampered by two shortcomings in current benchmarks.

With the widespread application of electroencephalography (EEG) in neuroscience and clinical practice, efficiently retrieving and semantically interpreting large-scale, multi-source, heterogeneous EEG data has become a pressing challenge. We propose EEG-MedRAG, a three-layer hypergraph-based retrieval-augmented generation framework that unifies EEG domain knowledge, individual patient cases, and a large-scale repository into a traversable n-ary relational hypergraph, enabling joint semantic-temporal retrieval and causal-chain diagnostic generation. Concurrently, we introduce the first cross-disease, cross-role EEG clinical QA benchmark, spanning seven disorders and five authentic clinical perspectives. This benchmark allows systematic evaluation of disease-agnostic generalization and role-aware contextual understanding. Experiments show that EEG-MedRAG significantly outperforms TimeRAG and HyperGraphRAG in answer accuracy and retrieval, highlighting its strong potential for real-world clinical decision support. Our data and code are publicly available at https://github.com/yi9206413-boop/EEG-MedRAG.

The limitations of unimodal deep learning models, particularly their tendency to overfit and limited generalizability, have renewed interest in multimodal fusion strategies. Multimodal deep neural networks (MDNN) have the capability of integrating diverse data domains and offer a promising solution for robust and accurate predictions. However, the optimal fusion strategy, intermediate fusion (feature-level) versus late fusion (decision-level) remains insufficiently examined, especially in high-stakes clinical contexts such as ECG-based cardiovascular disease (CVD) classification. This study investigates the comparative effectiveness of intermediate and late fusion strategies using ECG signals across three domains: time, frequency, and time-frequency. A series of experiments were conducted to identify the highest-performing fusion architecture. Results demonstrate that intermediate fusion consistently outperformed late fusion, achieving a peak accuracy of 97 percent, with Cohen's d > 0.8 relative to standalone models and d = 0.40 compared to late fusion. Interpretability analyses using saliency maps reveal that both models align with the discretized ECG signals. Statistical dependency between the discretized ECG signals and corresponding saliency maps for each class was confirmed using Mutual Information (MI). The proposed ECG domain-based multimodal model offers superior predictive capability and enhanced explainability, crucial attributes in medical AI applications, surpassing state-of-the-art models.

Research on applications of reinforcement learning (RL) to large language models has mostly been focused on single-turn problems, such as mathematical reasoning or single-shot code generation. While these problems can be viewed as token-level multi-turn Markov decision processes (MDPs), this view corresponds to a degenerate case of multi-turn interaction where the environment provides no feedback. This contrasts with many real-world domains, such as software engineering (SWE), which require rich multi-turn interactions with a stateful environment that responds to each action with a non-trivial observation. To bridge this gap, we demonstrate the successful application of RL to this general regime. Our methodology begins with rejection fine-tuning (RFT) using execution feedback to train a policy to follow instructions and formatting effectively, followed by a synchronous RL pipeline using DAPO for iterative improvement. Applying this pipeline to Qwen2.5-72B-Instruct, we increase its Pass@1 on the SWE-bench Verified benchmark from 11% to 39%, substantially improving upon the 20% RFT baseline. On the May and June splits of SWE-rebench, the resulting agent achieves Pass@1 of 35% and 31% respectively, competitive with even larger models such as DeepSeek-V3-0324 or Qwen3-235B-A22B, demonstrating that our methodology offers a practical approach for training capable agents for multi-turn interactive tasks using open-weight models.

The rapid proliferation of scientific knowledge presents a grand challenge: transforming this vast repository of information into an active engine for discovery, especially in high-stakes domains like healthcare. Current AI agents, however, are constrained by static, predefined strategies, limiting their ability to navigate the complex, evolving ecosystem of scientific research. This paper introduces HealthFlow, a self-evolving AI agent that overcomes this limitation through a novel meta-level evolution mechanism. HealthFlow autonomously refines its high-level problem-solving policies by distilling procedural successes and failures into a durable, structured knowledge base, enabling it to learn not just how to use tools, but how to strategize. To anchor our research and provide a community resource, we introduce EHRFlowBench, a new benchmark featuring complex health data analysis tasks systematically derived from peer-reviewed scientific literature. Our experiments demonstrate that HealthFlow's self-evolving approach significantly outperforms state-of-the-art agent frameworks. This work offers a new paradigm for intelligent systems that can learn to operationalize the procedural knowledge embedded in scientific content, marking a critical step toward more autonomous and effective AI for healthcare scientific discovery.