Search Results (388)

Accurate state estimation is essential for the real-time operation and control of modern distribution systems characterized by high renewable energy penetration, bidirectional power flows, and volatile loads. Conventional model-driven approaches such as the Weighted Least Squares (WLS) exhibit limited robustness under noisy and sparse measurements, while existing data-driven methods often neglect critical physical constraints inherent to power systems. To address these limitations, this paper proposes a physics-constrained Graph Attention Network (GAT) framework for distribution system state estimation (DSSE) that synergistically integrates data-driven learning with physical domain knowledge. The proposed method comprises three key components: (1) a Gaussian Mixture Model (GMM)-based data augmentation strategy that captures the stochastic characteristics of loads and distributed generation to generate synthetic samples consistent with actual operating distributions; (2) a GAT-based feature extractor with topology-aware admittance matrix embedding that effectively learns spatial dependencies and structural relationships among network nodes; and (3) a physics-constrained loss function that incorporates nodal power and voltage limit penalties to enforce operational feasibility. Comprehensive evaluations on the real-world 141-bus test system demonstrate that the proposed method achieves mean absolute error (MAE) reductions of 52.4% and 45.5% for voltage magnitude and angle estimation, respectively, compared to conventional Graph Convolutional Network (GCN)-based approaches. These results validate the superior accuracy, robustness, and adaptability of the proposed framework under challenging measurement conditions. Full article

Plant diseases are currently a major threat to agricultural economies and food availability, having a negative environmental impact. Despite being a promising line of research, current approaches struggle with poor cross-site generalization, limited labels and dataset bias. Real-field complexities, such as environmental variability, heterogeneous varieties or temporal dynamics as are often overlooked. Numerous studies have been conducted to address these challenges, proposing advanced learning strategies and improved evaluation protocols. Synthetic data generation and self-supervised learning reduce dataset bias, while domain adaptation, hyperspectral, and thermal signals improve robustness across sites. However, a large portion of current methods are developed and validated mainly on clean laboratory datasets, which do not capture the variability of real-field conditions. Existing AI models often lead to imperfect detection results when dealing with field images complexities, such as dense vegetation, variable illumination or changing symptom expression. Although augmentation techniques can approximate real-world conditions, incorporating field data represents a substantial enhancement in model reliability. Federated transfer learning offers a promising approach to enhance plant disease detection, by enabling collaborative training of models across diverse agricultural environments, using in-field data but without disclosing the participants data to each others. In this study, we collaboratively trained a hybrid Graph–SNN model using federated learning (FL) to preserve data privacy, optimized for efficient use of participant resources. The model achieved an accuracy of 0.9445 on clean laboratory data and 0.6202 exclusively on field data, underscoring the considerable challenges posed by real-world conditions. Our findings demonstrate the potential of FL for privacy preserving and reliable plant disease detection under real field conditions. Full article

Traditional Environmental, Social, and Governance (ESG) assessments rely heavily on corporate disclosures and third-party ratings, which are often delayed, inconsistent, and prone to bias. These limitations leave stakeholders without timely visibility into rapidly evolving ESG events. These assessment frameworks also fail to capture the dynamic nature of ESG issues reflected in public news media. This research addresses these limitations by proposing and implementing an automated framework utilising Artificial Intelligence (AI), specifically Natural Language Processing (NLP) and Knowledge Graphs (KG), to analyse ESG news data for companies listed on major stock indices. The methodology involves several stages: collecting a registry of target companies; retrieving relevant news articles; applying Named Entity Recognition (NER), sentiment analysis, and ESG domain classification; and constructing a linked property knowledge graph to structure the extracted information semantically. The framework culminates in an interactive dashboard for visualising and querying the resulting graph database. The resulting knowledge graph supports comparative inferential analytics across indices and sectors, uncovering divergent ESG sentiment profiles and thematic priorities that traditional reports overlook. The analysis also reveals comparative insights into sentiment trends and ESG focus areas across different exchanges and sectors, offering perspectives often missing from traditional methods. Findings indicate differing ESG sentiment profiles and thematic focuses between the UK (FTSE) and Australian (ASX) indices within the analysed dataset. This study confirms AI/KG’s potential for a modular, dynamic, and semantically rich ESG intelligence approach, transforming unstructured news into interconnected insights. Limitations and areas for future work, including model refinement and integration of financial data, are also discussed. This proposed framework augments traditional ESG evaluations with automated, scalable, and context-rich analysis. Full article

This paper presents a comprehensive review of motion capture (MoCap) data recovery techniques, with a particular focus on the suitability of artificial intelligence (AI) for addressing missing or corrupted motion data. Existing approaches are classified into three categories: non-data-driven, data-driven (AI-based), and hybrid methods. Within the AI domain, frameworks such as generative adversarial networks (GANs), transformers, and graph neural networks (GNNs) demonstrate strong capabilities in modeling complex spatial–temporal dependencies and achieving accurate motion reconstruction. Compared with traditional methods, AI techniques offer greater adaptability and precision, though they remain limited by high computational costs and dependence on large, high-quality datasets. Hybrid approaches that combine AI models with physics-based or statistical algorithms provide a balance between efficiency, interpretability, and robustness. The review also examines benchmark datasets, including CMU MoCap and Human3.6M, while highlighting the growing role of synthetic and augmented data in improving AI model generalization. Despite notable progress, the absence of standardized evaluation protocols and diverse real-world datasets continues to hinder generalization. Emerging trends point toward real-time AI-driven recovery, multimodal data fusion, and unified performance benchmarks. By integrating traditional, AI-based, and hybrid approaches into a coherent taxonomy, this review provides a unique contribution to the literature. Unlike prior surveys focused on prediction, denoising, pose estimation, or generative modeling, it treats MoCap recovery as a standalone problem. It further synthesizes comparative insights across datasets, evaluation metrics, movement representations, and common failure cases, offering a comprehensive foundation for advancing MoCap recovery research. Full article

This study presents a comparative evaluation of Dijkstra’s algorithm, A*, Bellman-Ford, AI-Augmented A* and a neural AI-based model for shortest-path computation across diverse graph topologies, with a focus on time efficiency and memory consumption under standardized experimental conditions. We analyzed grids, random graphs, and scale-free graphs of sizes up to 103,103 nodes, specifically examining 100- and 1000-node grids, 100- and 1000-node random graphs, and 100-node scale-free graphs. The algorithms were benchmarked through repeated runs per condition on a server-class system equipped with an Intel Xeon Gold 6248R processor, NVIDIA Tesla V100 GPU (32 GB), 256 GB RAM, and Ubuntu 20.04. A* consistently outperformed Dijkstra’s algorithm when paired with an informative admissible heuristic, exhibiting faster runtimes by approximately 1.37× to 1.91× across various topologies. In comparison, Bellman-Ford was slower than Dijkstra’s by approximately 1.50× to 1.92×, depending on graph type and size; however, it remained a robust option in scenarios involving negative edge weights or when early-termination conditions reduced practical iterations. The AI model demonstrated the slowest performance across conditions, incurring runtimes that were 2.60× to 3.23× higher than A* and 1.62× to 2.15× higher than Bellman-Ford, offering limited gains as a direct solver. These findings underscore topology-sensitive trade-offs: A* is preferred when a suitable heuristic is available; Dijkstra’s serves as a strong baseline in the absence of heuristics; Bellman-Ford is appropriate for handling negative weights; and current AI approaches are not yet competitive for exact shortest paths but may hold promise as learned heuristics to augment A*. We provide environmental details and comparative results to support reproducibility and facilitate further investigation into hybrid learned-classical strategies. Full article

Construction sites are complex environments where traditional safety monitoring methods often suffer from low detection accuracy and limited interpretability. To address these challenges, this study proposes a modular multimodal agent framework that integrates computer vision, knowledge representation, and large language model (LLM)–based reasoning. First, the CLIP model fine-tuned with Low-Rank Adaptation (LoRA) is combined with YOLOv10 to achieve precise recognition of construction activities and personal protective equipment (PPE). Second, a construction safety knowledge graph integrating Retrieval-Augmented Generation (RAG) is constructed to provide structured domain knowledge and enhance contextual understanding. Third, the FusedChain prompting strategy is designed to guide large language models (LLMs) to perform step-by-step safety risk reasoning. Experimental results show that the proposed approach achieves 97.35% accuracy in activity recognition, an average F1-score of 0.84 in PPE detection, and significantly higher performance than existing methods in hazard reasoning. The modular design also facilitates scalable integration with more advanced foundation models, indicating strong potential for real-world deployment in intelligent construction safety management. Full article

Predicting multiple future pedestrian trajectories is a challenging task for real-world applications like autonomous driving and robotic motion planning. Existing methods primarily focus on immediate spatial interactions among pedestrians, often overlooking the impact of distant spatial environments on their future trajectory choices. Additionally, aligning trajectory smoothness and temporal consistency remains challenging. We propose a multimodal trajectory prediction model that utilizes spatio-temporal graphical attention networks for crowd scenarios. Our method begins by generating simulated multiview pedestrian trajectory data using CARLA. It then combines original and selected multiview trajectories using a convex function to create augmented adversarial trajectories. This is followed by encoding pedestrian historical data with a multitarget detection and tracking algorithm. Using the augmented trajectories and encoded historical information as inputs, our spatio-temporal graph Transformer models scaled spatial interactions among pedestrians. We also integrate a trajectory smoothing method with a Memory Storage Module to predict multiple future paths based on historical crowd movement patterns. Extensive experiments demonstrate that our proposed MTP-STG model achieves state-of-the-art performance in predicting multiple future trajectories in crowds. Full article

Generating realistic synthetic gene expression data that captures the complex interdependencies and biological context of cellular systems remains a significant challenge. Existing methods often struggle to reproduce intricate co-expression patterns and incorporate prior biological knowledge effectively. To address these limitations, we propose BioGen-KI, a novel bio-inspired generative network with knowledge integration. Our framework leverages a hybrid deep learning architecture that integrates embeddings learned from biological knowledge graphs (e.g., gene regulatory networks, pathway databases) with a conditional generative adversarial network (cGAN). The knowledge graph embeddings guide the generator to produce synthetic expression profiles that respect known biological relationships, while conditioning on contextual information (e.g., cell type, experimental condition) allows for targeted data synthesis. Furthermore, we introduce a biologically informed discriminator that evaluates not only the statistical realism but also the biological plausibility of the generated data, encouraging the preservation of pathway coherence and relevant gene interactions. We demonstrate the efficacy of BioGen-KI by generating synthetic gene expression datasets that exhibit improved statistical similarity to real data and, critically, better preservation of biologically meaningful relationships compared to baseline GAN models and methods relying solely on statistical characteristics. Evaluation on downstream tasks, such as clustering and differential gene expression analysis, highlights the utility of BioGen-KI-generated data for enhancing the robustness and interpretability of biological data analysis. This work presents a significant step towards generating more biologically faithful synthetic gene expression data for research and development. Full article

The powerful graph convolutional network (GCN) for polarimetric synthetic aperture radar (PolSAR) image classification generally relies on real-valued features, ignoring the phase information and thus limiting the modeling of complex-valued (CV) polarization characteristics. To address this issue, this paper proposes a novel multiscale attention-enhanced CV graph U-Net model, abbreviated as MAE-CV-GUNet, by embedding CV-GCN into a graph U-Net framework augmented with multiscale attention mechanisms. First, a CV-GCN is constructed based on the real-valued GCN, to effectively capture the intrinsic amplitude and phase information of the PolSAR data, along with the underlying correlations between them. This way can well lead to an improved feature representation for PolSAR images. Based on CV-GCN, a CV graph U-Net (CV-GUNet) architecture is constructed by integrating multiple CV-GCN components, aiming to extract multi-scale features and further enhance the ability to extract discriminative features in the complex domain. Then, a multiscale attention (MSA) mechanism is designed, enabling the proposed MAE-CV-GUNet to adaptively learn the importances of features at various scales, thereby dynamically fusing the multiscale information among them. The comparisons and ablation experiments on three PolSAR datasets show that MAE-CV-GUNet has excellent performance in PolSAR image classification. Full article

As Large Language Models (LLMs) are core components in Retrieval-Augmented Generation (RAG) systems for knowledge-intensive tasks, concerns regarding hallucinations, redundancy, and unverifiable outputs have intensified, particularly in high-stakes domains, such as e-government. This study proposes a modular, multi-pipeline framework for statement-level faithfulness evaluation for characterizing hallucination and redundancy across both simple and agentic RAG pipelines. Using GPT-4.1, Claude Sonnet-4.0, and Gemini 2.5 Pro as LLM-based judges, this study examines how tool-specific attribution within agentic multi-tool architectures influences the interpretability and traceability of the generated content. By using a modular agentic RAG framework combining symbolic (GraphRAG), semantic (embedding), and real-time (web) retrieval, we benchmark hallucination and redundancy patterns, using state-of-the-art LLM judges. The study examines RAG and agent-based pipelines that attribute outputs to distinct tools, in contrast to traditional single-source RAG systems that rely on aggregated retrieval. Using e-government data sourced from the European Commission’s Press Corner, our evaluation framework assesses not only the frequency, but also the source-aware detectability of hallucinated content. The findings provide actionable insights into how source granularity and retrieval orchestration impact faithfulness evaluation across different pipeline architectures, while also suggesting new directions for explainability-aware RAG design. The study contributes a reproducible, modular framework for automated faithfulness assessment, with implications for transparency, governance compliance, and trustworthy AI deployment. Full article

Online learning environments generate vast amounts of student interaction data. While these records capture observable behaviors, they do not directly reveal students’ underlying knowledge states, which are essential for tracking learning progress. Knowledge tracing (KT) addresses this gap by predicting students’ future performance on exercises related to specific concepts, thereby enabling personalized learning and intelligent tutoring. Existing deep learning-based KT methods achieve promising results, but they often overemphasize either the sequential evolution of knowledge or the static structural relationships, which does not reflect the dynamic evolution of student learning. Moreover, they fail to model students’ knowledge state accurately under sparse interactions. To overcome these limitations, we propose DyGAS, a dynamic graph-augmented sequence modeling framework for knowledge tracing. The sequential module captures the dynamics pattern of knowledge acquisition and forgetting, while the structural module employs graph convolutional networks (GCN) to model inter-concept dependencies and knowledge transfer. Additionally, we propose that static knowledge modeling provides semantic priors to stabilize the representation of sparse concepts. Empirical results on three benchmark datasets demonstrate that DyGAS achieves superior performance compared to state-of-the-art methods, offering accurate and robust knowledge tracing across diverse learning scenarios. Full article

Entity resolution in real-world datasets remains a persistent challenge, particularly for identifying households and detecting co-residence patterns within noisy and incomplete data. While Large Language Models (LLMs) show promise, monolithic approaches often suffer from limited scalability and interpretability. This study introduces a multi-agent Retrieval-Augmented Generation (RAG) framework that decomposes household entity resolution into coordinated, task-specialized agents implemented using LangGraph. The system includes four agents responsible for direct matching, transitive linkage, household clustering, and residential movement detection, combining rule-based preprocessing with LLM-guided reasoning. Evaluation on synthetic S12PX dataset segments containing 200–300 records demonstrates 94.3% accuracy on name variation matching and a 61% reduction in API calls compared to single-LLM baselines, while maintaining transparent and traceable decision processes. These results indicate that coordinated multi-agent specialization improves efficiency and interpretability, providing a structured and extensible approach for entity resolution in census, healthcare, and other administrative data domains. Full article

English argumentative writing is a cornerstone of academic and professional communication, yet it remains a significant challenge for second-language (L2) learners. While Large Language Models (LLMs) show promise as components in automated feedback systems, their responses are often generic and lack the structural insight necessary for meaningful improvement. Existing Automated Essay Scoring (AES) systems, conversely, typically provide holistic scores without the kind of actionable, fine-grained advice that can guide concrete revisions. To bridge this systemic gap, we introduce ARGUS (Argument Understanding and Structured-feedback), a novel neuro-symbolic system that synergizes the semantic understanding of LLMs with the structured reasoning of Graph Neural Networks (GNNs). The ARGUS system architecture comprises three integrated modules: (1) an LLM-based parser transforms an essay into a structured argument graph; (2) a Relational Graph Convolutional Network (R-GCN) analyzes this symbolic structure to identify specific logical and structural flaws; and (3) this flaw analysis directly guides a conditional LLM to generate feedback that is not only contextually relevant but also pinpoints precise weaknesses in the student’s reasoning. We evaluate ARGUS on the Argument Annotated Essays corpus and on an additional set of 150 L2 persuasive essays collected from the same population to augment training of both the parser and the structural flaw detector. Our argument parsing module achieves a component identification F1-score of 90.4% and a relation identification F1-score of 86.1%. The R-GCN-based structural flaw detector attains a macro-averaged F1-score of 0.83 across the seven flaw categories, indicating that the enriched training data substantially improves its generalization. Most importantly, in a human evaluation study, feedback generated by the ARGUS system was rated as consistently and significantly more specific, accurate, actionable, and helpful than that from strong baselines, including a fine-tuned LLM and a zero-shot GPT-4. Our work demonstrates a robust systems engineering approach, grounding LLM-based feedback in GNN-driven structural analysis to create an intelligent teaching system that provides targeted, pedagogically valuable guidance for L2 student writers engaging with persuasive essays. Full article

Drug discovery remains a time-consuming and costly process, necessitating innovative computational approaches to accelerate early stage target identification and compound development. We introduce AgentMol, a modular multimodel AI system that integrates large language models, chemical language modeling, and deep learning–based affinity prediction to automate the discovery pipeline. AgentMol begins with disease-related queries processed through a Retrieval-Augmented Generation system using the Large Language Model to identify protein targets. Protein sequences are then used to condition a GPT-2–based chemical language model, which generates corresponding small-molecule candidates in SMILES format. Finally, a regression convolutional neural network (RCNN) predicts the drug-target interaction by estimating binding affinities (pKi). Models were trained and validated on 470,560 ligand–protein pairs from the BindingDB database. The chemical language model achieved high validity (1.00), uniqueness (0.96), and diversity (0.89), whereas the RCNN model demonstrated robust predictive performance with R² > 0.6 and Pearson’s R > 0.8. By leveraging LangGraph for orchestration, AgentMol delivers a scalable, interpretable pipeline, effectively enabling the end-to-end generation and evaluation of drug candidates conditioned on protein targets. This system represents a significant step toward practical AI-driven molecular discovery with accessible computational demands. Full article

Despite the success of reinforcement learning (RL), the common assumption of online interaction prevents its widespread adoption. Offline RL has emerged as an alternative that learns a policy from precollected data. However, this learning paradigm introduces a new challenge called “distributional shift”, degrading the performance of the policy when evaluated on out-of-distribution (OOD) scenarios (i.e., outside of the training data). Most existing works resolve this by policy regularization to optimize a policy within the support of the data. However, this overlooks the potential for high-reward regions outside of the data. This motivates offline policy optimization that is capable of finding high-reward regions outside of the data. In this paper, we devise a causality-based model architecture to accurately capture the OOD scenarios wherein the policy can be optimized without performance degradation. Specifically, we adapt causal normalizing flows (CNFs) to learn the transition dynamics and reward function for data generation and augmentation in offline policy learning. Based on the physics-based qualitative causal graph and precollected data, we develop a model-based offline OOD-adapting causal RL (MOOD-CRL) algorithm to learn the quantitative structural causal model. Consequently, MOOD-CRL can exercise counterfactual reasoning for sequential decision-making, revealing a high potential for OOD adaptation. The effectiveness is validated through extensive empirical evaluations with ablations including data quality and algorithmic sensitivity. Our results show that MOOD-CRL achieves comparable results with its online counterparts and consistently outperforms state-of-the-art model-free and model-based baselines by a significant margin. Full article