Blockchain technology and cryptocurrencies have attracted significant attention in recent years, yet remain susceptible to cyber threats such as phishing attacks. Existing detection approaches often suffer from high computational costs and limited robustness, especially when facing varying data distributions and sparse structures. To address these issues, we propose Robust, Node behavior, Transaction structure, and Network (R-NTN), a detection framework for Ethereum phishing accounts that leverages multi-dimensional transaction features. R-NTN first constructs 2-hop ego graphs via random walks, then extracts features from three complementary dimensions: behavioral attributes, transaction-based structural features, and network embeddings. These features are integrated into a unified representation for downstream classification. Experiments show that R-NTN consistently outperforms baseline methods and maintains high accuracy across datasets of different scales and compositions, demonstrating strong robustness and generalizability.

Background/Objectives: Autism spectrum disorder (ASD) exhibits remarkable genetic heterogeneity involving hundreds of risk genes; however, the mechanism by which these genes organize within biological networks to contribute to disease pathogenesis remains incompletely understood. This study aims to elucidate these organizational principles and identify critical network bottlenecks using a novel integrative computational framework. Methods: We analyzed 893 SFARI genes using a three-pronged computational approach: (1) a Machine Learning Dynamic Perturbation Propagation algorithm; (2) a hypergraph construction method explicitly modeling multi-gene complexes by integrating protein-protein interactions, co-expression modules, and curated pathways; and (3) Hypergraph Neural Network embeddings for gene clustering. Validation was performed using hub-independent features to address potential circularity, followed by a druggability assessment to prioritize therapeutic targets. Results: The hypergraph construction captured 3847 multi-way relationships, representing a 45% increase in biological relationships compared to pairwise networks. The perturbation algorithm achieved a 51% higher correlation with TADA genetic evidence than random walk methods. Analysis revealed a hierarchical organization where 179 hub genes exhibited a 3.22-fold increase in degree centrality and a 4.71-fold increase in perturbation scores relative to non-hub genes. Hypergraph Neural Network clustering identified five distinct gene clusters, including a "super-hub" cluster of 10 genes enriched in synaptic signaling (4.2-fold) and chromatin remodeling (3.9-fold). Validation confirmed that 8 of these 10 genes co-cluster even without topological information. Finally, we identified high-priority therapeutic targets, including ARID1A, POLR2A, and CACNB1. Conclusions: These findings establish hierarchical network organization principles in ASD, demonstrating that hub genes maintain substantially elevated perturbation states. The identification of critical network bottlenecks and pharmacologically tractable targets provides a foundation for understanding autism pathogenesis and developing precision medicine approaches.

LightDTA: lightweight drug-target affinity prediction via random-walk network embedding and knowledge distillation.

The convergence of Artificial Intelligence (AI) and the Internet of Things (IoT) is enabling a new class of intelligent applications. Specifically, Large Language Models (LLMs) are emerging as powerful tools not only for natural language understanding but also for enhancing IoT security. However, the integration of these computationally intensive models into resource-constrained IoT environments presents significant challenges. This paper provides an in-depth examination of how LLMs can be adapted to secure IoT ecosystems. We identify key application areas, discuss major challenges, and propose optimization strategies for resource-limited settings. Our primary contribution is a novel collaborative embeddings offloading mechanism for IoT intrusion detection named SEED (Semantic Embeddings for Efficient Detection). This system leverages a lightweight, fine-tuned BERT model, chosen for its proven contextual and semantic understanding of sequences, to generate rich network embeddings at the edge. A compact neural network deployed on the end-device then queries these embeddings to assess network flow normality. This architecture alleviates the computational burden of running a full transformer on the device while capitalizing on its analytical performance. Our optimized BERT model is reduced by approximately 90% from its original size, now representing approximately 41 MB, suitable for the Edge. The resulting compact neural network is a mere 137 KB, appropriate for the IoT devices. This system achieves 99.9% detection accuracy with an average inference time of under 70 ms on a standard CPU. Finally, the paper discusses the ethical implications of LLM-IoT integration and evaluates the resilience of LLMs in dynamic and adversarial environments.

ABSTRACT This article presents a novel integration of two theoretical concepts traditionally examined and applied separately in policing scholarship: feminist ethics of care and organisational justice. Drawing from pioneering empirical work on online harms management in UK police organisations, we analyse 52 interviews with managerial personnel from four partner forces to problematise the emphasis on organisational reputation in online harms management. By applying a feminist ethics of care lens, we theorise and demonstrate how care-based principles, such as attentiveness, relationality, and contextual responsiveness, can underpin organisationally just management strategies in practice. We propose embedding networks of care structures and dependencies within police organisations and advancing a politics of care to enhance perceptions of organisational justice among police personnel, while ensuring managerial officers are supported in and through their care relations. This theoretical integration offers new visions and actionable approaches to care which empowers police personnel and potentially translates into wider public good through more democratic policing. Our policy-oriented recommendations for online harms management are generalisable to international policing contexts and other public-facing professions beyond policing.

Automated community detection of biomedical composite networks across network embedding and dynamic optimization.

Multimodal hazardous materials risk graph completion: a joint optimization approach with dual channel embedding and generative adversarial network

A novel dual-knowledge embedded graph convolutional network method combining knowledge quantification for gas turbine gas path analysis

A time-frequency graph fusion framework for Major Depressive Disorder diagnosis in multi-site rsfMRI data.

The paper introduces a unified method of designing sculptures on a feeling-sensitive neural network basis. The proposed Emotion-Form Neural Embedding Network (EFNEN) is based on the combination of Convolutional Neural Networks (CNNs) and Graph Neural Networks (GNNs) to learn emotion-related correlations between sculptural form and emotion. The system was trained and tested using a selection of 1,200 annotated 3D models that had both geometric and a set of 3 emotion labels (valence and arousal) assigned to them. EFNEN obtained a correlation coefficient (r = 0.88) and 92.4% accuracy with human perceptual ratings, which was better than the baseline models. Latent emotion space and feature-emotion heatmap visualizations showed that the predictors of positive affect are curvature, symmetry, and balance. The model facilitates the classification of emotions as well as emotion-driven three dimensional form generation, thus leading to collaborative co-creation of artists and AI systems. The findings indicate that emotion is calculally formulated and synthesized to form a measurable aesthetic dimension, which makes EFNEN a platform of affective computational art and human-AI creative synergy.

ABSTRACT Environmental, Social, and Governance (ESG) performance has become a crucial factor in determining a firm's resilience and strategic agility in the evolving capital markets environment. This study examines how ESG practices contribute to organizational resilience through the mediating role of dynamic capabilities and the moderating effect of social network embedding, operationalized via board directors' social capital (centrality and structural holes). The proposed framework is tested using moderated mediation models on panel data from 7560 A‐share companies listed on the Shanghai and Shenzhen stock markets from 2013 to 2022. The findings demonstrate three major findings. First, ESG performance significantly enhances firms' dynamic capabilities, which include innovation, adaptation, and absorptive capacity, thereby strengthening their resilience to external shocks. Second, a stronger board network position amplifies the ESG‐capability link. Third, heterogeneity analyses reveal that the ESG‐resilience nexus is more pronounced in eastern regions, among large‐scale firms, non‐state‐owned enterprises, and in post‐pandemic periods. According to the study, ESG is a strategic resource. It highlights the role of director networks in the process of utilizing ESG investments to promote the long‐term sustainability of organizations. These results can contribute to the body of research on dynamic capabilities and stakeholder governance by providing a cross‐level model that incorporates sustainability, network theory, and resilience in emerging market environments.

Mainstream research in audio-visual learning has focused on designing task-specific expert models, primarily implemented through sophisticated multimodal fusion approaches. Recently, a few efforts have aimed to develop more task-independent or universal audiovisual embedding networks, encoding advanced representations for use in various audiovisual downstream tasks. This is typically achieved by fine-tuning large pretrained transformers, such as Swin-V2-L and HTS-AT, in a parameter-efficient manner through techniques such as tuning only a few adapter layers inserted into the pretrained transformer backbone. Although these methods are parameter-efficient, they suffer from significant training memory consumption due to gradient backpropagation through the deep transformer backbones, which limits accessibility for researchers with constrained computational resources. In this paper, we present Meta-Token Learning (Mettle), a simple and memory-efficient method for adapting large-scale pretrained transformer models to downstream audio-visual tasks. Instead of sequentially modifying the output feature distribution of the transformer backbone, Mettle utilizes a lightweight Layer-Centric Distillation (LCD) module to distill in parallel the intact audio or visual features embedded by each transformer layer into compact meta-tokens. This distillation process considers both pretrained knowledge preservation and task-specific adaptation. The obtained meta-tokens can be directly applied to classification tasks, such as audio-visual event localization and audio-visual video parsing. To further support fine-grained segmentation tasks, such as audio-visual segmentation, we introduce a Meta-Token Injection (MTI) module, which utilizes the audio and visual meta-tokens distilled from the top transformer layer to guide feature adaptation in earlier layers. Extensive experiments on multiple audiovisual benchmarks demonstrate that our method significantly reduces memory usage and training time while maintaining parameter efficiency and competitive accuracy.

Single-cell RNA sequencing (scRNA-seq) enables high-resolution characterization of cellular heterogeneity, but its rich, complementary structure across cells and genes remains underexploited, especially in the presence of technical noise and sparsity. Effectively leveraging this multi-scale structure is essentially an information fusion problem that requires integrating heterogeneous graph-based views of cells and genes into robust low-dimensional representations. In this paper, we introduceGatorSC, a unified representation learning framework that models scRNA-seq data through multi-scale cell and gene graphs and fuses them with a Mixture-of-Experts architecture.GatorSCconstructs a global cell–cell graph, a global gene–gene graph, and a local gene–gene graph derived from neighborhood-specific subgraphs, and learns graph neural network embeddings that are adaptively fused by a gating network. To learn noise-robust and structure-preserving embeddings without labels, we couple graph reconstruction and graph contrastive learning in a unified self-supervised objective applied to both cell- and gene-level graphs. We evaluateGatorSCon 19 publicly available scRNA-seq datasets covering diverse tissues, species, and sequencing platforms. Across 14 benchmark datasets,GatorSCconsistently outperforms state-of-the-art deep generative, graph-based, and contrastive methods for cell clustering, gene expression imputation, and cell-type annotation. The learned embeddings are used for accurate trajectory inference, recovery of canonical marker gene programs, and cell-type-specific pathway signatures in an Alzheimer’s disease singlenucleus dataset.GatorSCprovides a flexible foundation for comprehensive single-cell transcriptomic analysis and can be readily extended to multi-omic and spatial modalities.

Environmental sustainability and firm performance: Unpacking the productivity effects of green data center pilots.

Virtual Network Embedding: Literature Assessment, Recent Advancements, Opportunities, and Challenges

Artificial Intelligence (AI) has revolutionized modern medicine by providing computational solutions to manage complex clinical data and improve therapeutic outcomes. In pharmacology, AI particularly machine learning (ML) and deep learning (DL) models has demonstrated significant potential in predicting drug–drug interactions (DDIs), a major cause of adverse drug reactions (ADRs) and increased healthcare costs. This study focuses on the DANN-DDI (Deep Attention Neural Network for Drug–Drug Interaction) model, which integrates diverse pharmacological data to enhance the accuracy of DDI prediction. Drug features including chemical substructures, targets, enzymes, pathways, and existing interactions were extracted from the DrugBank (version 5.1.0) and KEGG databases. These features were used to construct five drug-feature networks, and structural deep network embedding (SDNE) was employed to learn drug representations. The DANN-DDI framework consists of three components: drug feature learning, drug-pair feature learning, and interaction prediction using a deep neural network optimized via the Adam algorithm and binary cross-entropy loss. Model performance was evaluated using 5-fold cross-validation and assessed through AUC, AUPR, accuracy, and F-measure metrics. The results indicated that optimal parameters (embedding dimension = 128, 7 hidden layers, 150 epochs, dropout rate = 0.4) yielded superior prediction outcomes. Compared with traditional computational methods such as similarity analysis and matrix factorization, the DANN-DDI model demonstrated improved capability to detect potential DDIs effectively. Overall, this study highlights the value of integrating AI-based approaches into pharmacovigilance systems to predict and prevent harmful drug interactions, ultimately enhancing patient safety and treatment efficacy.

Abstract The recent progress of using graph based encoding of crystal structures for high throughput material property prediction has been quite successful. However, using a single modality model prevents us from exploiting the advantages of an enhanced features space by combining different representations. Specifically, pre-trained Large language models(LLMs) can encode a large amount of knowledge which is beneficial for training of models. Moreover, the graph encoder is able to learn the local features while the text encoder is able to learn global information such as space group and crystal symmetry. In this work, we propose Material Multi-Modal Fusion(MatMMFuse), a fusion based model which uses a multi-head attention mechanism for the combination of structure aware embedding from the Crystal Graph Convolution Network (CGCNN) and text embeddings from the SciBERT model. We train our model in an end-to-end framework using data from the Materials Project Dataset. We show that our proposed model shows an improvement compared to the vanilla CGCNN and SciBERT model for all four key properties- formation energy, band gap, energy above hull and fermi energy. Specifically, we observe an improvement of 40% compared to the vanilla CGCNN model and 68% compared to the SciBERT model for predicting the formation energy per atom. Importantly, we demonstrate the zero shot performance of the trained model on small curated datasets of Perovskites, Chalcogenides and the Jarvis Dataset. The results show that the proposed model exhibits better zero shot performance than the individual plain vanilla CGCNN and SciBERT model. This enables researchers to deploy the model for specialized industrial applications where collection of training data is
prohibitively expensive.

Despite the underdeveloped formal institutional system in China’s capital market, the venture capital (VC) industry has continued to grow rapidly, exhibiting a clear trend of network formation. To better understand the formation of VC networks, this study systematically analyzes factors from three dimensions: endogenous network structures, multidimensional relational networks among VC firms, and informal networks of venture capitalists. Using data from the Wind database and other sources, networks are constructed based on 1317 investment events involving 157 VC firms. An exponential random graph model is applied to assess the effects of multiple network embeddings on VC network formation. The results reveal that, among endogenous structural factors, triad closure structures are more likely to be embedded in VC networks than two-path structures with brokerage functions. In terms of exogenous factors, the geographic distance network among VC firms exerts a negative effect on VC network formation, while knowledge proximity networks—i.e., those based on industry, investment stage, and region—positively influence VC networks formation. Informal networks of venture capitalists increase the probability of VC network formation. Compared with previous studies, this research is based on self-organization, market-oriented, and relational logics, integrating multiple factors—including endogenous network structures, venture capital firm characteristics, and venture capitalists—and introduces a cross-network perspective to build a novel multilevel network embedding ERGM framework to examine VC network formation. Furthermore, the study reveals how informal ties substitute for formal institutions in China’s VC network formation.

Joint-node-link Mapping for Virtual Network Embedding in 5G/6G Environments Using GAT-augmented PPO

Abstract Context: Over the past few years, there has been a growing trend in the utilization of network embedding techniques for predicting software defects. Network Embeddings capture extensive information about software networks, but not all embeddings are equally pertinent for defect prediction. The presence of irrelevant and redundant embeddings has adversely affected the complexity and performance of software defect prediction (SDP) models. Objective: In the pursuit of optimizing defect prediction, the objective of this work is twofold: (i) utilizing network embeddings extracted from call graphs to identify latent and complex features that capture intricate class relationships, (ii) applying feature selection techniques to identify defect prediction-relevant network embeddings and addressing class imbalance through data balancing techniques for developing an SDP model. Method: This study utilizes 10 software projects, employing 6 different network embedding algorithms to extract 32 and 128-dimensional embeddings from each project’s call graph. Seven feature selection techniques are evaluated by applying each of them to a comprehensive set of 250 datasets. SMOTE is applied to datasets for enhancing training fairness and predictive accuracy. The effectiveness of these techniques in SDP is assessed by developing models using 22 different classifiers. Performance metrics, including accuracy and AUC, are evaluated, while cost-effectiveness is also considered. A threshold is established based on testing efficiency and defect removal cost. Result: Through the application of feature selection methods and utilizing a smaller set of selected embeddings, the proposed SDP model achieved a mean AUC value of 72%, demonstrating an improvement over models that incorporated all available embeddings. The combination of embeddings and software metrics outperformed software metrics and embeddings by 3% in terms of AUC. Following feature selection, the 128-dimensional embeddings displayed nearly the same level of performance as the 32-dimensional embeddings. SMOTE application yielded notable performance improvements on highly imbalanced datasets. Conclusion: The result shows that the rank sum feature selection technique consistently highlights its effectiveness when compared to other feature selection methods. The proposed SDP framework has the ability to exhibit performance capabilities similar to those achieved when using lower-dimensional embeddings, indicating the superiority of these simplified models that use a lesser number of embeddings while still containing a rich set of software component relationships compared to existing techniques. Also, SMOTE effectively addressed the dataset imbalance, enhancing defect prediction performance on imbalanced datasets.