IEEE transactions on artificial intelligence最新文献

{"title":"Talk Structurally, Act Hierarchically: A Collaborative Refinement Framework for LLM Multi-Agent Systems","authors":"Zhao Wang;Sota Moriyama;Wei-Yao Wang;Briti Gangopadhyay;Shingo Takamatsu","doi":"10.1109/TAI.2026.3676025","DOIUrl":"https://doi.org/10.1109/TAI.2026.3676025","url":null,"abstract":"While recent advancements in large language model-based multi-agent (LLM-MA) systems have shown promise, significant challenges remain in managing communication and refinement when agents collaborate on complex tasks. In this article, we propose <italic>talk structurally, act hierarchically (TalkHier)</i>, a novel framework that introduces a structured communication protocol for context-rich exchanges and a hierarchical refinement system to address issues such as incorrect outputs, falsehoods, and biases. <italic>TalkHier</i> surpasses various types of SoTA, including inference scaling model (OpenAI-o1), open-source multiagent models (e.g., AgentVerse), and majority voting strategies on current LLM and single-agent baselines (e.g., ReAct and GPT4o), across diverse tasks, including open-domain question answering, domain-specific selective questioning, and practical advertisement text generation. These results highlight its potential to set a new standard for LLM-MA systems, paving the way for more effective, adaptable, and collaborative multiagent frameworks.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4993-5003"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859724","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":"Dual-Fault Tolerant Control for Connected Vehicular Platoons Based on Extended State Fuzzy Observation","authors":"Ruifeng Wang;Peihao Du;Yan Lei;Qi Zhou;Tingwen Huang","doi":"10.1109/TAI.2026.3685962","DOIUrl":"https://doi.org/10.1109/TAI.2026.3685962","url":null,"abstract":"This article investigates an adaptive fuzzy active fault-tolerant control problem for connected automated vehicular platoons operating under actuator faults and inter-vehicle ranging sensor faults. A saturated Nussbaum function is employed to handle actuator faults, suppressing the initial acceleration oscillations caused by conventional Nussbaum-based control and thereby improving the transient response of the vehicular platoon. Simultaneously, within the control framework, an adaptive compensation scheme is constructed to restore the reliability of inter-vehicle distance measurements compromised by onboard ranging sensor faults. In addition, an extended state fuzzy observer is designed to estimate both the system states and external disturbances, thereby overcoming the difficulty of obtaining reliable and accurate vehicle motion information in the presence of internal dynamic uncertainties and external disturbances. Finally, the stability of the vehicular platoon can be verified by Lyapunov stability criteria, and sufficient validations on eight vehicles are implemented for validating the efficacy of the proposed control approach.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5255-5267"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859703","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":"ASBench: Image Anomalies Synthesis Benchmark for Anomaly Detection","authors":"Qunyi Zhang;Songan Zhang;Jiaqi Liu;Jinbao Wang;Xiaoning Lei;Guoyang Xie;Guannan Jiang;Zhichao Lu","doi":"10.1109/TAI.2026.3680823","DOIUrl":"https://doi.org/10.1109/TAI.2026.3680823","url":null,"abstract":"Anomaly detection plays a pivotal role in manufacturing quality control, yet its application is constrained by limited abnormal samples and high manual annotation costs. While anomaly synthesis offers a promising solution, existing studies predominantly treat anomaly synthesis as an auxiliary component within anomaly detection frameworks, lacking systematic evaluation of anomaly synthesis algorithms. Current research also overlooks crucial factors specific to anomaly synthesis, such as decoupling its impact from detection, quantitative analysis of synthetic data and adaptability across different scenarios. To address these limitations, we propose ASBench, the first comprehensive benchmarking framework dedicated to evaluating anomaly synthesis methods. Our framework introduces four critical evaluation dimensions: 1) the generalization performance across different datasets and pipelines; 2) the ratio of synthetic to real data; 3) the correlation between intrinsic metrics of synthesis images and anomaly detection performance metrics; and 4) strategies for hybrid anomaly synthesis methods. Through extensive experiments, ASBench not only reveals limitations in current anomaly synthesis methods but also provides actionable insights for future research directions in anomaly synthesis.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5162-5176"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859739","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":"Dual Adversarial Adaptation With Dynamic Labeling Mechanism for Semisupervised Cross-Well Lithology Identification","authors":"Ji Chang;Jing Li;Tao Shen;Dewen Liang;Yunbo Zhao;Shaoqun Dong;Yu Kang;Wenjun Lv","doi":"10.1109/TAI.2026.3679484","DOIUrl":"https://doi.org/10.1109/TAI.2026.3679484","url":null,"abstract":"Well log instruments continuously measure multiple geophysical parameters of wellbore rocks, providing high-resolution observation data for identifying subsurface lithology. Recently, lithology identification methods based on semisupervised learning and transfer learning have been investigated to address the widespread issue of label scarcity. However, these methods only account for the distribution discrepancy between wells (domains), while ignoring the data divergence within a well. This can result in inaccurate pseudolabels within a well and erroneous label matching across wells. To cope with these challenges, we propose a dual adversarial adaptation method with dynamic labeling mechanism for semisupervised cross-well lithology identification. Our method leverages intradomain adversarial learning to promote fine-grained category alignment and interdomain adversarial learning to generate domain-invariant feature representations, thereby reducing the data differences within and between wells. Additionally, our method introduces a dynamic label adjustment mechanism to optimize the selection of target pseudolabels and source samples, thereby increasing reliable supervision information and improving the feature alignment process. Experimental results on actual well log datasets demonstrate that our method can achieve effective cross-well lithology identification when target labels are scarce.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5149-5161"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859749","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":"On Convergence of Adam With Data Dependent Stepsize","authors":"Alokendu Mazumder;Rishab Sabharwal;Bhartendu Kumar;Manan Tayal;Chirag Garg;Arnab Roy;Punit Rathore","doi":"10.1109/TAI.2026.3656866","DOIUrl":"https://doi.org/10.1109/TAI.2026.3656866","url":null,"abstract":"In neural network training, Adam is a widely favored optimization algorithm. A crucial factor in their performance is selecting the correct stepsize, significantly impacting their effectiveness. The stepsize is a crucial hyperparameter that requires careful tuning to achieve optimal results. For training large models on large-scale datasets, tuning the stepsize can be very expensive and often time-consuming. Stepsize schedulers are proposed to navigate the trajectory on highly non-convex losses in deep learning, but many decaying schedulers (such as linear decay, exponential decay, and step decay) can lead to divergence or slow convergence compared to constant stepsizes. This suggests that rapidly decaying rates play a crucial role in driving convergence. Building on this observation, we propose a new estimate of constant stepsize that depends on the dynamics of the network and the data. This approach eliminates two major issues: the problems associated with decaying schedulers and the need for an exhaustive search for stepsizes. Numerical experiments demonstrate that our estimated stepsize closely matches the best-performing stepsizes found via exhaustive search, significantly reducing the need for manual tuning. We also show that under certain conditions, our stepsize estimator performs on par with various decaying schedulers. We further conduct experiments on the full ImageNet dataset to demonstrate the effectiveness of our proposed method in a large-scale setting. Beyond image classification, we also apply our approach to large language model (LLM) fine-tuning, benchmarking it against a range of modern optimizers. These results highlight the versatility of our method across diverse domains. Furthermore, we prove that with our estimated stepsize, Adam can reach critical points for smooth, non-convex objectives within bounded runtime, providing strong theoretical support for our approach. We analyze both deterministic and stochastic versions of Adam and establish sufficient conditions for our proposed stepsize to achieve asymptotic convergence of the gradients to zero. Our results also show that the key driver for Adam’s convergence is the use of non-increasing stepsizes, despite accumulating a few past gradients.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4806-4821"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859730","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":"Multimodal Transformer-Based Gait Analysis and Deep Learning Model for Slip-Resistant Footwear Evaluation","authors":"Shaghayegh Chavoshian;Atena Roshan Fekr","doi":"10.1109/TAI.2026.3681277","DOIUrl":"https://doi.org/10.1109/TAI.2026.3681277","url":null,"abstract":"Slips and falls remain a leading cause of injury, particularly in icy winter conditions and occupational settings with slippery surfaces. Several laboratory and field-based systems are designed to assess footwear slip resistance, including mechanical testers, tribometers, and human-centered gait-based assessments. While mechanical tests offer standardized slip resistance metrics for footwear, they often fail to reflect real-world variability in human gait, environmental conditions, and anthropometric diversity. On the other hand, although human-centered testing ensures ecological validity, it typically reports a single aggregated score for footwear derived from multiple participants, which obscures individual biomechanical differences and limits the ability to provide personalized assessments of slip resistance performance. This study introduces a transformer-based regression model that integrates: 1) temporal biomechanical features; 2) outsole images; 3) footwear metadata; and 4) anthropometric data to predict personalized slip resistance score of 84 footwear across 578 walking trials conducted by 18 human participants on two types of ice: wet and dry. The predictive performance of our multimodal deep learning model varied by surface condition, with substantially better results on dry surfaces (RMSE: 1.34<inline-formula><tex-math>${}^mathbf{circ}$</tex-math></inline-formula> <inline-formula><tex-math>$mathbf{pm}$</tex-math></inline-formula> 0.34<inline-formula><tex-math>${}^mathbf{circ}$</tex-math></inline-formula>, R<inline-formula><tex-math>${}^mathbf{2}$</tex-math></inline-formula>: 0.45 <inline-formula><tex-math>$mathbf{pm}$</tex-math></inline-formula> 0.09, MAPE: 10.48% <inline-formula><tex-math>$mathbf{pm}$</tex-math></inline-formula> 3.58%) than on wet surfaces (RMSE: 2.34<inline-formula><tex-math>${}^mathbf{circ}$</tex-math></inline-formula> <inline-formula><tex-math>$mathbf{pm}$</tex-math></inline-formula> 0.43<inline-formula><tex-math>${}^mathbf{circ}$</tex-math></inline-formula>, R<inline-formula><tex-math>${}^mathbf{2}$</tex-math></inline-formula>: 0.47 <inline-formula><tex-math>$mathbf{pm}$</tex-math></inline-formula> 0.12, MAPE: 21.45% <inline-formula><tex-math>$mathbf{pm}$</tex-math></inline-formula> 6.89%). The results demonstrate the potential of fusing biomechanical and outsole characteristics to reliably assess slip risk under varying environmental conditions. Additionally, they emphasize the need for adaptive and context-aware testing protocols and pave the way for the development of intelligent footwear assessment tools aimed at injury prevention.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5189-5198"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11475610","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859737","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":"Generalized Holographic Reduced Representations","authors":"Calvin Yeung;Zhuowen Zou;SungHeon Jeong;Wenjun Huang;Nathaniel D. Bastian;Mohsen Imani","doi":"10.1109/TAI.2026.3678232","DOIUrl":"https://doi.org/10.1109/TAI.2026.3678232","url":null,"abstract":"Hyperdimensional computing (HDC) is a computationally and data-efficient paradigm that acts as a bridge between connectionist and symbolic approaches to artificial intelligence (AI). However, HDC’s simplicity poses challenges for encoding complex compositional structures, especially in its binding operation. To address this, we propose generalized holographic reduced representations (GHRRs), an extension of Fourier holographic reduced representations (FHRRs), a specific HDC implementation. GHRR introduces a flexible, noncommutative binding operation, enabling improved encoding of complex data structures while preserving HDC’s desirable properties of robustness and transparency. In this work, we introduce the GHRR framework, prove its theoretical properties and its adherence to HDC properties, explore its kernel and binding characteristics, and perform empirical experiments showcasing its flexible noncommutativity, enhanced decoding accuracy for compositional structures. We also demonstrate that binding in GHRR is more expressive than that in other HDC variants; in particular, we show that binding in GHRR can implement a kind of attention mechanism. We verify this by replacing the attention mechanism in a transformer with its GHRR-equivalent and testing it on a language modeling task, showing improved performance compared to a vanilla transformer.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5061-5075"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11456911","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859743","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":"Graph Attention Transformer With Multitask Learning for Motion Prediction in Autonomous Driving","authors":"Chuan Hu;Hao Jiang;Peichuan Lang;Biao Yang;Xiaobo Chen;Hao Chen","doi":"10.1109/TAI.2026.3672361","DOIUrl":"https://doi.org/10.1109/TAI.2026.3672361","url":null,"abstract":"The motion prediction of surrounding vehicles is crucial for autonomous driving. As it can effectively guide the planning and decision-making of the autonomous vehicle. Although many methods have been proposed to predict the future motion of vehicles, but they failed capture the essense of multitask learning for vehicle motion prediction and lacked effective semantic and interaction information fusion. In this article, we propose a novel model, graph attention transformer with multitask learning, to predict the future motion of the target vehicle by modeling motion prediction task as a multitask learning problem of trajectory prediction and driving intention estimation. The model consists of three modules: feature fusion, spatio-temporal interaction, and motion decoder. The feature fusion module is used to effectively fuse the semantic and dynamic information of the target vehicle. The Spatial Interaction module is used to model the interaction between vehicles. The motion decoder module decouples the motion prediction into multitask learning for trajectory prediction and driving intention estimation in a parallel manner. Extensive experiments on real-world datasets, with average improvement of 50% in NGSIM and 40% in HighD, demonstrate that the proposed model achieves superior performance.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4893-4905"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859717","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":"Heterophily-Based Dynamic Hypergraph for Semisupervised Classification and Its Application in UAV Fault Diagnosis","authors":"Shaojun Liang;Ying Zheng;Housheng Su","doi":"10.1109/TAI.2026.3672585","DOIUrl":"https://doi.org/10.1109/TAI.2026.3672585","url":null,"abstract":"Hypergraph neural network (HGNN) excels at representing intricate relationships within non-Euclidean infinite-dimensional spaces. However, traditional hypergraph frameworks are grounded in homophily assumption and encounter challenges in handling heterophily problem in semisupervised classification tasks. To tackle this problem, this article first analyzes the underlying reasons that limit homophily-based hypergraphs, and proposes a heterophily-based dynamic structure. Under this structure, a heterophily-based dynamic HGNN (HDHGNN) is then proposed for semisupervised classification. HDHGNN decomposes the hypergraph’s triple pattern iterative process into two causally related subprocedures, within which attention mechanisms are incorporated to enable dynamic interactions. To evaluate the effectiveness of HDHGNN, a relative entropy-based heterophily degree evaluation method is first introduced, followed by comprehensive fault classification experiments on two real-world flight datasets from rotary-wing and fixed-wing unmanned aerial vehicles, with comparisons against 11 state-of-the-art methods.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4906-4917"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859732","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":"Cross-Corpus Speech Emotion Recognition Based on Dynamically Filtering Multistage Diffusion Model","authors":"Jingjie Yan;Boyan Sun;Yuebo Yue;Xiaoyang Zhou;Ying Liu","doi":"10.1109/TAI.2026.3681265","DOIUrl":"https://doi.org/10.1109/TAI.2026.3681265","url":null,"abstract":"In cross-corpus speech emotion recognition (SER) tasks, although generative methods have shown certain potential in alleviating the scarcity of target domain samples, the generated samples often lack precise alignment with the acoustic-semantic features of the target domain. This results in pseudosamples of uneven quality and even introduces noise, thereby weakening the effectiveness of transfer learning. To address the problem of insufficient quality control over pseudosamples in existing methods, we propose a dynamically filtering multistage diffusion model (DFMDM). The proposed DFMDM model establishes a three-stage iterative optimization process of “generation–filtering–enhancement.” In the first stage, the conditional diffusion model is guided by an adversarial domain classifier to generate pseudosamples with target domain characteristics. In the second stage, a dynamic filtering module is presented to eliminate low-quality pseudosamples that significantly deviate from the target domain features. In the third stage, the filtered pseudosamples are mixed with source domain samples and reused to regenerate target-style pseudosamples. In this way, a progressive diffusion-based enhancement loop is constructed, which improves sample quality through iterative generation and dynamic adjustment of similarity thresholds. Experimental results conducted on CASIA, EmoDB, eNTERFACE, and IEMOCAP demonstrate that the proposed DFMDM model achieves significant improvements over existing state-of-the-art (SOTA) approaches, fully showcasing its application potential and practical value in cross-corpus SER.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5177-5188"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859736","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}