IEEE transactions on artificial intelligence最新文献

{"title":"LearningFlow: Automated Policy Learning Workflow for Urban Driving With Large Language Models","authors":"Zengqi Peng;Yubin Wang;Xu Han;Lei Zheng;Jun Ma","doi":"10.1109/TAI.2026.3686819","DOIUrl":"https://doi.org/10.1109/TAI.2026.3686819","url":null,"abstract":"Recent advancements in reinforcement learning (RL) demonstrate the significant potential in autonomous driving. Despite this promise, challenges such as the manual design of reward functions and low sample efficiency in complex environments continue to impede the development of safe and effective driving policies. To tackle these issues, we introduce LearningFlow, an innovative automated policy learning workflow tailored to urban driving. This framework leverages the collaboration of multiple large language model (LLM) agents throughout the RL training process. LearningFlow includes a curriculum sequence generation process and a reward generation process, which work in tandem to guide the RL policy by generating tailored training curricula and reward functions. Particularly, each process is supported by an analysis agent that evaluates training progress and provides critical insights to the generation agent. Through the collaborative efforts of these LLM agents, LearningFlow automates policy learning across a series of complex driving tasks, and it significantly reduces the reliance on manual reward function design while enhancing sample efficiency. Comprehensive experiments are conducted in the high-fidelity CARLA simulator, along with comparisons with other existing methods, to demonstrate the efficacy of our proposed approach. The results demonstrate that LearningFlow excels in generating rewards and curricula. It also achieves superior performance and robust generalization across various driving tasks, as well as commendable adaptation to different RL algorithms.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5429-5444"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097410","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":"Adversary-Robust Graph-Based Learning of WSIs","authors":"Saba Heidari Gheshlaghi;Milan Aryal;Nasim Yahyasoltani;Masoud Ganji","doi":"10.1109/TAI.2026.3688616","DOIUrl":"https://doi.org/10.1109/TAI.2026.3688616","url":null,"abstract":"Enhancing the robustness of deep learning (DL) models against adversarial attacks is crucial, particularly in critical domains such as healthcare, where significant financial interests increase the risk of such attacks. Whole slide images (WSIs) are high-resolution, digitized tissue samples used for cancer diagnosis; however, their large size and multiresolution nature make them vulnerable to both image- and graph-level attacks. In this work, we propose a graph-based defense model that utilizes a graph neural network (GNN) to extract features from graph representations of WSIs while integrating a denoising module based on untrained neural network prior (UNNP) and graph signal processing principles, along with a vision transformer (ViT) for adversarial defense. The denoising module leverages overparameterized networks and early stopping to filter out adversarial perturbations, ensuring improved robustness by optimizing the GNN parameters for each input graph independently rather than relying on a pretrained model. This reduces the impact of adversarial noise at both the image level, addressing attacks on WSIs stored in the cloud, and the graph level, countering perturbations introduced during GNN training. Additionally, the ViT module enhances prostate cancer Gleason grading by capturing complex dependencies in the data. We evaluated the model on clean WSI samples, with and without the denoiser, to establish a baseline and assess its effectiveness. The model was then tested against various adversarial attacks, demonstrating a significant improvement in cancer diagnosis accuracy. These results highlight the robustness and efficiency of the proposed approach in mitigating adversarial challenges in medical imaging.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 10","pages":"5563-5572"},"PeriodicalIF":0.0,"publicationDate":"2026-10-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"149097565","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":"Generalized-Hukuhara Subgradient-Based Neurodynamic Approach to Nonsmooth Constrained Distributed Interval-Valued Optimization","authors":"Jingxin Liu;Shuwen Liu;Yingxu Wang;Jun Peng;Xiaofeng Liao;Chunhong Du;Hangjun Che;Jin-Song Dong","doi":"10.1109/TAI.2026.3675069","DOIUrl":"https://doi.org/10.1109/TAI.2026.3675069","url":null,"abstract":"In this article, we propose a projection-type neurodynamic approach for solving a class of nonsmooth constrained distributed optimization problems with interval-valued local objective functions over a network of vertices. We introduce the concept of generalized-Hukuhara subgradient (<italic>gH</i>-subgradient) and differential inclusion system in the algorithm framework, aiming to cope with the confusion that objective functions may not be <italic>gH</i>-differentiable, and further deriving the efficiency condition of the solution of the problem by employing the geometry relationship between the <italic>gH</i>-subgradient and the normal cone. The boundedness and convergence of the state solution of the proposed neurodynamic algorithm are provable, and the conclusion that the state outputs of all vertices reach consensus on the efficient solution of the problem can be guaranteed by Lyapunov function method and Byrnes–Martin integral invariant principle. The designed neurodynamic model is simple in structure, avoids the calculation of matrix inversion and the introduction of Lagrange multipliers, and is suitable for solving general distributed optimization problems with interval-valued functions and constraints. Finally, the simulation applications of network utility, power flow balancing, and gas transportation show the viability of the proposed algorithm.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4963-4976"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859722","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":"Redefining Non-IID Data in Federated Learning for Computer Vision Tasks: Migrating From Labels to Embeddings for Task-Specific Data Distributions","authors":"Kasra Borazjani;Payam Abdisarabshali;Naji Khosravan;Seyyedali Hosseinalipour","doi":"10.1109/TAI.2026.3677838","DOIUrl":"https://doi.org/10.1109/TAI.2026.3677838","url":null,"abstract":"Federated learning (FL) has emerged as one of the prominent paradigms for distributed machine learning (ML). However, it is well-established that its performance can degrade significantly under non-IID (non-independent and identically distributed) data distributions across clients. To study this effect, the existing works predominantly emulate data heterogeneity by imposing label distribution skew across clients. In this article, we show that label distribution skew fails to fully capture the data heterogeneity in computer vision tasks beyond classification, <italic>exposing an overlooked gap in the literature</i>. Motivated by this, by utilizing pretrained deep neural networks to extract task-specific data embeddings, we define <italic>task-specific data heterogeneity</i> through the lens of each vision task and introduce a new level of data heterogeneity called <italic>embedding-based data heterogeneity</i>. Our methodology involves clustering data points based on embeddings and distributing them among clients using the Dirichlet distribution. Through extensive experiments, we evaluate the performance of different FL methods under our revamped notion of data heterogeneity, introducing new benchmark performance measures to the literature. For instance, across seven representative computer vision tasks, our embedding-based heterogeneity formulation leads to up to around 60% increase in the observed loss under FedAvg, indicating that it more accurately exposes the performance degradation caused by data heterogeneity. We further unveil a series of open research directions that can be pursued.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5045-5060"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859741","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":"Deep Reinforcement Learning With Adaptive Enhanced Attention for CTSPs","authors":"Renyi Zhang;Xiangping Xu;Xinli Shi;Jinde Cao;Jie Gui","doi":"10.1109/TAI.2026.3673681","DOIUrl":"https://doi.org/10.1109/TAI.2026.3673681","url":null,"abstract":"The colored traveling salesman problem (CTSP) and its extended variant, the cumulative capacitated colored traveling salesman problem (C²-CTSP), represent significant generalizations of the classical multiple TSP. By incorporating color constraints, capacity restrictions, and customized cost calculation mechanisms for different entities, these formulations demonstrate strong relevance to real-world applications. This article proposes an adaptive enhanced attention (AEA) model to address color-constrained traveling salesman problems based on a reinforcement learning framework. Our method introduces several novel techniques. First, we develop a dynamic color-aware decoding mechanism that systematically handles color constraints through bitwise operations and adaptive state transitions. Then, we establish a bi-objective optimization method integrating an adaptive penalty mechanism to balance cumulative cost minimization with constraint satisfaction. In addition, we design an enhanced attention unit (EAU) that accelerates strategy convergence and improves solution precision through gated activation and hierarchical feature propagation. Evaluations on CTSP and C²-CTSP benchmarks demonstrate that the AEA model outperforms existing neural combinatorial optimization solvers and several heuristic baselines. Generalization experiments reveal that the EAU module effectively accelerates convergence and enhances decoding robustness, showing promising applicability to other combinatorial optimization tasks.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4739-4750"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859713","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":"Human-in-the-Loop Formation-Containment Safe Control for Multi-agent Systems via Reinforcement Learning","authors":"Luning Yang;Pei Chi;Jiang Zhao;Yingxun Wang","doi":"10.1109/TAI.2025.3559040","DOIUrl":"https://doi.org/10.1109/TAI.2025.3559040","url":null,"abstract":"This article designs the optimal formation containment safe tracking control protocol for nonlinear second-order multi-agent systems (MASs) with unknown dynamics. A human operator can interact with the tracking leader of MASs to improve security and communicate with other agents through the MASs network in a complex environment. Primarily, adaptive neural networks are utilized to identify the unknown nonlinear dynamics parameter and ensure that the identification state errors converge asymptotically. Additionally, the optimized safe control design incorporates an actor–critic architecture based on reinforcement learning (RL) to approximate the Hamilton–Jacobi–Bellman (HJB) equation for nonlinear MASs. This distributed security control of MASs achieves optimal tracking using only local connectivity information. Finally, the proposed optimal algorithm for MASs safe formation-containment control is verified with MASs in an unknown obstacle environment.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4792-4805"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859714","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":"Budget-Constrained Online Retrieval-Augmented Generation: The Chunk-as-a-Service Model","authors":"Shawqi Al-Maliki;Ammar Gharaibeh;Mohamed Rahouti;Mohammad Ruhul Amin;Mohamed Abdallah;Junaid Qadir;Ala Al-Fuqaha","doi":"10.1109/TAI.2026.3666170","DOIUrl":"https://doi.org/10.1109/TAI.2026.3666170","url":null,"abstract":"Large language models (LLMs) have revolutionized the field of natural language processing. However, they exhibit some limitations, including a lack of reliability and transparency: they may hallucinate and fail to provide sources that support the generated output. Retrieval-augmented generation (RAG) was introduced to address such limitations in LLMs. One popular implementation, RAG-as-a-service (RaaS), has shortcomings that hinder its adoption and accessibility. For instance, RaaS pricing is based on the number of submitted prompts, without considering whether the prompts are enriched by relevant chunks, i.e., text segments retrieved from a vector database, or the quality of the utilized chunks (i.e., their degree of relevance). This results in an opaque and less cost-effective payment model. We propose chunk-as-a-service (CaaS) as a transparent and cost-effective alternative. CaaS includes two variants: open-budget CaaS (OB-CaaS) and limited-budget CaaS (LB-CaaS), which is enabled by our “utility-cost online selection algorithm (UCOSA).” UCOSA further extends the cost-effectiveness and the accessibility of the OB-CaaS variant by enriching, in an online manner, a subset of the submitted prompts based on budget constraints and utility-cost tradeoff. Our experiments demonstrate the efficacy of the proposed UCOSA compared to both offline and relevance-greedy selection baselines. In terms of the performance metric—the number of enriched prompts (<inline-formula><tex-math>$rm NEPs$</tex-math></inline-formula>) multiplied by the average relevance (<inline-formula><tex-math>$rm AR$</tex-math></inline-formula>)—UCOSA outperforms random selection by approximately 52% and achieves around 75% of the performance of offline selection methods. Additionally, in terms of budget utilization, LB-CaaS and OB-CaaS achieve higher performance-to-budget ratios of 140% and 86%, respectively, compared to RaaS, indicating their superior efficiency.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4822-4836"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11400550","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859715","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":"Efficient Out-of-Distribution Generalization for Pretrained GNNs via Prompt Learning","authors":"Zhenmeng Zuo;Changsheng Li;Mingkui Tan;Xing Gong;Ye Yuan;Guoren Wang","doi":"10.1109/TAI.2026.3666523","DOIUrl":"https://doi.org/10.1109/TAI.2026.3666523","url":null,"abstract":"Graph out-of-distribution (OOD) generalization enables graph neural networks (GNNs) to perform robustly in applications when confronted with new environments not encountered during training. For better realization of OOD generalization, current methods are increasingly focused on designing more complex causal inference models trained from scratch, without considering the potential of existing pretrained models, and increased computational complexity. In this article, we propose an efficient framework [efficient graph OOD generalization (EGOG)], which uses prompt learning for pretrained GNNs to extract causal subgraphs from input graph data, for enhancing the generalization ability of the pretrained GNNs. Specifically, we design a graph causal prompt module to enhance the generalization ability of the pretrained GNNs. First, we use graph prompt learning with a causal features generator that enables pretrained GNNs to decouple causal features from spurious features in graph nodes. Then, considering that the structure of graph is also affected by distribution shift, we further design a causal structure generator (CSG), which reuses the pretrained GNNs to analyze graph’s structure, decoupling causal substructures from spurious substructures. This approach enhances the pretrained GNN’s ability to decouple causal subgraph from both the node and structure perspectives. Meantime, we employ gradient reversal layers and information bottleneck to encourage our method to more accurately disentangle causal subgraph from spurious subgraph. Experiment results show that EGOG can be applied to pretrained GNNs to enable generalization ability. Compared with the state-of-the-art baseline, EGOG shows an average relative improvement of 2.73% on two synthetic datasets while using only average 4.38% of the trainable parameters.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5268-5277"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859731","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":"Nearest Neighbor Projection Removal Adversarial Training","authors":"Himanshu Singh;A. V. Subramanyam;Shivank Rajput;Mohan Kankanhalli","doi":"10.1109/TAI.2026.3681719","DOIUrl":"https://doi.org/10.1109/TAI.2026.3681719","url":null,"abstract":"Deep neural networks have exhibited impressive performance in image classification tasks but remain vulnerable to adversarial examples. Standard adversarial training enhances robustness but typically fails to explicitly address inter-class feature overlap, a significant contributor to adversarial susceptibility. In this work, we introduce a novel adversarial training framework that actively mitigates inter-class proximity by projecting out inter-class dependencies from adversarial and clean samples in the feature space. Specifically, our approach first identifies the nearest inter-class neighbors for each adversarial sample and subsequently removes projections onto these neighbors to enforce stronger feature separability. Theoretically, we demonstrate that our proposed logits correction reduces the Lipschitz constant of neural networks, thereby lowering the Rademacher complexity, which directly contributes to improved generalization and robustness. Extensive experiments across standard benchmarks including CIFAR-10, CIFAR-100, SVHN, and TinyImagenet show that our method demonstrates strong performance that is competitive with leading adversarial training techniques, highlighting significant achievements in both robust and clean accuracy. Our findings reveal the importance of addressing inter-class feature proximity explicitly to bolster adversarial robustness in deep neural networks (DNNs).","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5287-5295"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859734","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":"Tunable Domain Adaptation Using Unfolding","authors":"Snehaa Reddy;Jayaprakash Katual;Satish Mulleti","doi":"10.1109/TAI.2026.3679328","DOIUrl":"https://doi.org/10.1109/TAI.2026.3679328","url":null,"abstract":"Machine learning models often struggle to generalize across domains with varying data distributions, such as differing noise levels, leading to degraded performance. Traditional strategies such as personalized training, which trains separate models per domain, and joint training, which uses a single model for all domains, have significant limitations in flexibility and effectiveness. To address this, we propose two novel domain adaptation methods for regression tasks based on interpretable unrolled networks—deep architectures inspired by iterative optimization algorithms. These models leverage the functional dependence of select tunable parameters on domain variables, enabling controlled adaptation during inference. Our methods include parametric tunable-domain adaptation, which uses known domain parameters for dynamic tuning, and data-driven tunable-domain adaptation, which infers domain adaptation directly from input data. We validate our approach on compressed sensing problems involving noise-adaptive sparse signal recovery, domain-adaptive gain calibration, and domain-adaptive phase retrieval, demonstrating improved or comparable performance to domain-specific models while surpassing joint training baselines. This work highlights the potential of unrolled networks for effective, interpretable domain adaptation in regression settings.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5133-5148"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859735","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}