IEEE transactions on artificial intelligence最新文献

{"title":"IEEE Transactions on Artificial Intelligence Publication Information","authors":"","doi":"10.1109/TAI.2026.3733835","DOIUrl":"https://doi.org/10.1109/TAI.2026.3733835","url":null,"abstract":"","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"C2-C2"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11719382","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097323","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"CECO-LAD: Cloud-Edge Collaboration for Unsupervised Log Anomaly Detection","authors":"Qinxuan Shi;Zhanglong Yang;Sicong Shao","doi":"10.1109/TAI.2026.3701348","DOIUrl":"https://doi.org/10.1109/TAI.2026.3701348","url":null,"abstract":"Artificial intelligence (AI)-driven log anomaly detection (LAD) is a critical component for maintaining the security and reliability of cyber infrastructure. However, deploying an effective LAD system in real-world environments presents a significant challenge, the cloud-edge dilemma, where accurate deep learning models favor centralized cloud resources, but operational constraints (e.g., latency, bandwidth, privacy, and energy) favor edge-local analysis. To address these challenges, we propose a comprehensive collaborative framework for heterogeneous cloud-edge LAD (CECO-LAD) that balances detection accuracy with resource efficiency. In CECO-LAD, we use expectation-maximization-enhanced anomaly transformer (EM-AT) as the base learner for the novel ensemble learning model, Bagging-style AT (BAT), the cloud-side ensemble detector, and further develop quantized BAT (Q-BAT) for resource-constrained edge environments. A Mahalanobis distance-based routing policy enables cloud-edge collaboration by selectively forwarding only uncertain samples to the cloud and retaining confident cases at the edge, thereby minimizing resource consumption while maximizing detection accuracy. Additionally, we propose Green-LADE, a Green AI-inspired method to enable holistic edge-side evaluation. Experiments on the Hadoop Distributed File System (HDFS), OpenStack, and blue gene/l (BGL) datasets show that CECO-LAD achieves state-of-the-art anomaly detection performance while substantially reducing resource consumption on resource-constrained edge devices. CECO-LAD achieves <inline-formula><tex-math>$F_{1}$</tex-math></inline-formula>-scores of 99.98% on HDFS, 99.95% on OpenStack, and 99.99% on BGL. The proposed framework advances LAD by making it accurate, deployable, and sustainable.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5813-5829"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097490","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"LGAT: Toward Local-Global Adaptive Temporal Graph Representation Learning","authors":"Hongjiang Chen;Pengfei Jiao;Xuan Guo;Mengzhou Gao;Zhidong Zhao","doi":"10.1109/TAI.2026.3685609","DOIUrl":"https://doi.org/10.1109/TAI.2026.3685609","url":null,"abstract":"Temporal graph representation learning aims to model evolving networks by embedding nodes with respect to their dynamic interactions and attributes over time. Real-world temporal graphs typically exhibit power-law degree distributions, where a small subset of high-degree nodes dominates interactions while the majority of low-degree nodes maintain sparse connections. This inherent imbalance, characterized by node heterogeneity and edge sparsity, poses fundamental challenges: existing approaches that rely on the first-order neighborhood aggregation cannot adequately represent under-connected nodes, and the limited number of observed edges restricts the availability of supervision signals. To address these challenges, we propose local-global adaptive temporal graph representation learning (LGAT), which integrates local topological features with global evolutionary semantics. The core insight is that local neighborhood information alone is insufficient for heterogeneous networks, while a purely global perspective risks over-smoothing well-connected nodes. LGAT therefore adaptively balances these two perspectives based on each node’s structural role. Specifically, LGAT employs a hybrid temporal decay mechanism that combines fixed exponential decay with dynamic time decay to model temporal dependencies, quantifies mutual neighbor cooccurrence frequencies to capture fine-grained local topology, and maintains a global evolutionary representation that disproportionately benefits under-represented nodes through inverse-degree weighting. Furthermore, LGAT introduces a contrastive pretext task that enriches supervision signals by maximizing mutual information between interacting node pairs, directly mitigating the edge sparsity problem. Extensive experiments on six real-world datasets demonstrate that LGAT significantly outperforms state-of-the-art methods, with particularly pronounced improvements on networks exhibiting strong heterogeneity and sparsity.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5370-5382"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097471","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"EEG-Based Speech Imagery Classification via Neural Architecture Search","authors":"Taveena Lotey;Aman Verma;Rajkumar Saini;Partha Pratim Roy","doi":"10.1109/TAI.2026.3700595","DOIUrl":"https://doi.org/10.1109/TAI.2026.3700595","url":null,"abstract":"Speech imagery (SI) has emerged as a promising paradigm for brain–computer interfaces (BCIs) that enable hands-free communication without external stimuli. Classification of electroencephalography-based SI electroencephalography signals (SI-EEG) remains challenging due to limited data and high inter- and intra-subject variability. Most existing approaches rely on manually designed architectures, which is a time-consuming and labor-intensive process that requires significant domain expertise. To address these challenges, we propose a novel neural architecture search (NAS) framework for SI-EEG signal classification with bilevel optimization and path-level binarization to automatically and adaptively discover optimal model architectures. Furthermore, we propose <italic>Deformable ConvNeXt block</i>, a novel convolutional neural network-based operation that integrates transformer-inspired design principles and EEG-specific priors. This block enables each neuron to adaptively control the size of the EEG signal’s receptive field. It allows the model to more effectively capture temporal variations in mental activity in inter- and intra-subject classification settings. Evaluated across three SI-EEG datasets, our approach achieves classification accuracies of 68.30%, 68.49%, and 72.32%, surpassing the state of the art. These results demonstrate the effectiveness and scalability of our NAS-based framework for next-generation real-time BCI systems.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5790-5802"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097475","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"A Collaborative Neural Network With Adaptive Gradient Dynamics for Nonconvex Optimization","authors":"Juliang Wang;Haoen Huang;Kun Deng;Zhigang Zeng","doi":"10.1109/TAI.2026.3695501","DOIUrl":"https://doi.org/10.1109/TAI.2026.3695501","url":null,"abstract":"In this article, a collaborative neural network (CONN) with adaptive gradient dynamics is presented for addressing nonconvex optimization problems. Existing neural dynamics models often suffer from either prohibitive computational costs or unsatisfactory performance, which limits their practical effectiveness. To overcome this, an adaptive gradient neural dynamics (AGND) model is developed for local optimization in nonconvex problems, achieving reduced computational complexity and superior performance. Subsequently, a CONN is proposed that combines multiple AGND models for scattered searches with a metaheuristic rule that reinitializes neuronal states upon local convergence. Theoretical analysis establishes that the AGND model ensures local optimality, and further demonstrates the global convergence of the proposed CONN. Extensive experiments on numerous classical time-invariant nonconvex optimization benchmark problems illustrate the effectiveness and stability of the CONN. Moreover, the proposed method consistently achieves strong performance across diverse application problems, including both time-varying and time-invariant scenarios, such as robotic manipulator trajectory tracking, portfolio selection, and pressure vessel design.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5613-5626"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097568","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Autonomous Pipeline Inspection Using Lightweight Vision-Based UAV Tracking","authors":"Naisarg Pandya;Surya Prakash S.K.;Ayush Gupta;Ashok Kumar Sivarathri;Jagadeesh Rachapudi;Pranav Singh;Praful Hambarde;Amit Shukla","doi":"10.1109/TAI.2026.3693978","DOIUrl":"https://doi.org/10.1109/TAI.2026.3693978","url":null,"abstract":"Long-range pipelines are indispensable infrastructures requiring continuous and rigorous inspections for structural health monitoring. Multicopter unmanned aerial vehicle (UAV) inspections face critical challenges including low-altitude flight constraints, limited computational resources for integrated visual perception and navigation control for the purpose of continuous inspection in real-time. The present work proposes an autonomous vision-based UAV low altitude inspection framework for industrial pipeline inspection. The proposed framework seamlessly integrates three innovative components: a lightweight deep learning architecture (MLA-Net) for robust pipeline segmentation, a vision-based control strategy for smooth tracking, and a sliding mode controller for real-time navigation. MLA-Net combines cascaded feature extractor (CFE) blocks for hierarchical feature learning with multi-scale receptive (MSR) blocks for multi-scale environmental adaption, achieving efficient pipe detection under varying conditions. The vision-based control strategy processes MLA-Net outputs through perspective projection and oriented rectangle extraction to coordinate yaw-lateral control, while a nonlinear forward velocity function prevents overshooting during pipeline turns. The sliding mode controller fuses visual features with altitude data for autonomous navigation across complex pipeline networks. Extensive validation through simulation and real-time outdoor experiments demonstrates superior performance across diverse pipeline configurations including intersections and multipipe scenarios. The proposed MLA-Net achieves 0.8166 IoU while operating at <inline-formula><tex-math>$approx$</tex-math></inline-formula>27 FPS with only 0.202 GFLOPs and 0.178 M, significantly outperforming existing methods in computational efficiency. The integrated framework enables stable autonomous tracking across various pipeline geometries, providing a practical solution with enhanced accuracy and computational efficiency for resource-constrained UAV applications. Source code available at Autonomous vision-based pipline tracking by UAV.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5597-5612"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097659","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"SF-DynFinGC: Scale-Free Dynamic Graph Construction for Event-Driven Spatiotemporal Learning in Financial Anomaly Detection","authors":"Yifan Chen;Haiqi Zhu;Haoxuan Xu;Chunzhi Yi;Zhiyuan Chen;Dario Landa-Silva","doi":"10.1109/TAI.2026.3698049","DOIUrl":"https://doi.org/10.1109/TAI.2026.3698049","url":null,"abstract":"Existing machine learning methods for financial anomaly detection (FAD) typically overlook the complex spatiotemporal interactions inherent in financial data, significantly limiting detection accuracy and interpretability. While graph-based approaches have shown promise in modeling relational structures, existing methods mainly rely on static or overly simplistic dynamic graph constructions, struggling to capture these complex spatiotemporal interactions. To address these limitations, we propose SF-DynFinGC, a novel dynamic graph construction framework specifically tailored for event-driven spatiotemporal learning in FAD. SF-DynFinGC integrates scale-free structural regularization with a decay-governed global Top-k edge selection strategy, effectively reflecting real-world hierarchical and dispersed financial network characteristics. Through dynamically evolving graph structures triggered by financial events rather than fixed intervals, SF-DynFinGC accommodates irregular temporal and spatial data distributions. Comprehensive experiments on various benchmark datasets demonstrate that SF-DynFinGC consistently achieves state-of-the-art performance across graph neural network (GNN) and traditional classifiers, significantly outperforming existing static and dynamic graph construction approaches. Structural analyses further reveal that SF-DynFinGC generates interpretable and appropriately sparse graphs, highlighting critical transactional relationships indicative of complex anomaly patterns. This study provides a robust and interpretable dynamic graph construction framework that enhances both detection accuracy and analytical transparency, thereby extending the practical applicability of graph-based methods in financial risk management.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5693-5706"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097779","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Tumor Image Segmentation Using Multimodal Generation and Vision Transformer Integration","authors":"Yizhan Fan;Xiren Zhou;Zhenchao Tao;Huanhuan Chen","doi":"10.1109/TAI.2026.3699617","DOIUrl":"https://doi.org/10.1109/TAI.2026.3699617","url":null,"abstract":"Accurate tumor segmentation is crucial for clinical diagnosis and radiotherapy planning. However, computed tomography (CT)-only or magnetic resonance imaging (MRI)-only imaging often fails to capture the full tumor morphology due to modality-specific limitations. To address these limitations and leverage the complementary strengths of CT and MRI, we propose multimodal generative polygon-vision transformer (<sc>MGP-</small>ViT), which integrates CT-to-MRI generative synthesis with a polygon-based boundary representation to enhance tumor boundary fidelity. Specifically, a cycle-consistent adversarial network (CycleGAN)-based CT-to-MRI translator generates synthetic MRI scans from unpaired CT data, improving soft-tissue contrast without requiring spatially paired CT–MRI training samples. These synthetic MRI and original CT scans are input to a dual-branch vision transformer (ViT) encoder: each branch extracts modality-specific features, which are fused via a modality-aware cross-attention module and then decoded into polygon vertex sequences that outline tumor regions. We further introduce a convolutional smoothing loss (CSL) that encourages locally smooth vertex trajectories by penalizing large, incoherent jumps while tolerating small spatial offsets. Experiments on the public MSD pancreas dataset and a private cervical-cancer CT cohort show that <sc>MGP-ViT</small> outperforms state-of-the-art methods in terms of intersection over union (IoU) and dice, while offering improved boundary precision, greater robustness to imaging variability, and better cross-dataset generalization.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5764-5775"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097956","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Predator–Prey Model: Driven Hunt for Accelerated Grokking","authors":"I. A. Lopatin;S. V. Kozyrev;Alexander N. Pechen","doi":"10.1109/TAI.2026.3699326","DOIUrl":"https://doi.org/10.1109/TAI.2026.3699326","url":null,"abstract":"A machine learning method is proposed using two agents (representing two coupled replicas of the model parameters) that simulate the biological behavior of a predator and a prey. In this method, the two agents interact with each other via a pair of potentials that simulate the dynamics when the predator chases the prey while the prey runs away from the predator—to perform an optimization on the landscape. This method allows, for optimization landscapes with narrow ravines with gentle slope or even flat bottoms, to avoid optimization getting stuck in a ravine. For some examples of grokking (i.e., delayed generalization) problems we show that this method allows for achieving up to a hundred times faster learning in gradient evaluations and up to 60 times in the wall-clock time compared to the standard learning procedure.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5753-5763"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097551","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Structural Node Representation on Irregular Roles","authors":"Wang Zhang;Wenjun Wang;Xuan Guo;Tianpeng Li;Pengfei Jiao;Lin Pan;Jian Wu;Xue Chen","doi":"10.1109/TAI.2026.3698801","DOIUrl":"https://doi.org/10.1109/TAI.2026.3698801","url":null,"abstract":"Roles of nodes refer to the structural patterns of relations that reflect node identities or functions in the graph or network. Learning role-based node representations has been a hot topic in the last decade, as they are helpful in fully understanding the internal mechanisms of the network and improving performance on downstream tasks. In real-world networks, the dependencies between the roles are extremely complex, where the local structures of nodes can vary considerably even if they share the same role. However, existing methods ignore the varying dependencies and fail to discover roles with an irregular structure. To fully understand the above problems, we first propose an index based on a graph kernel to measure role irregularity and analyze its influence on downstream tasks. Then, for practically overcoming the shortcomings, we propose a novel framework based on a graph neural network and a variational auto-encoder with a cross-attention mechanism (GVCAM). It leverages the graph convolutional network to preserve the varying dependencies between nodes and the variational auto-encoder to capture the structural traits of nodes. We introduce the cross-attention mechanism that fuses the two parts to adaptively learn the irregular roles. Extensive experiments on one synthetic and six real-world datasets show that GVCAM achieves competitive performance across multiple tasks compared with ten state-of-the-art baselines.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5707-5721"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097720","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}