IEEE transactions on artificial intelligence最新文献_第6页

Fine-Grained Features in Any-Shot Learning: A Survey 任意镜头学习中的细粒度特征：综述

IEEE transactions on artificial intelligence Pub Date : 2026-09-01 Epub Date: 2026-03-31 DOI: 10.1109/TAI.2026.3678885

Qiangchang Wang;Wenhao Li;Jinglin Zhang;Zheyun Qin;Linhao Li

{"title":"Fine-Grained Features in Any-Shot Learning: A Survey","authors":"Qiangchang Wang;Wenhao Li;Jinglin Zhang;Zheyun Qin;Linhao Li","doi":"10.1109/TAI.2026.3678885","DOIUrl":"https://doi.org/10.1109/TAI.2026.3678885","url":null,"abstract":"Deep learning has achieved remarkable success across a wide range of domains and applications over the past few decades. However, the high performance of most deep learning models is heavily dependent on large volumes of labeled data, which is often labor-intensive, time-consuming, and sometimes impractical to collect. To address this challenge, any-shot learning (ASL)—such as few-shot and zero-shot learning—has been proposed to recognize novel classes with only a few labeled examples or class descriptions. The human visual system excels at identifying discriminative regions in complex scenes, and inspired by this capability, fine-grained features are extracted from input data to emphasize key regions, advancing the field of ASL. Despite the success of these methods, there is a lack of comprehensive surveys on the use of fine-grained features in ASL. This article provides a detailed survey of recent research on fine-grained features in few-shot and zero-shot learning, offering summaries of various approaches, discussing their strengths and limitations, and identifying key challenges, future directions, and research opportunities to drive further advancements in the field. We also explore the potential of ASL in semisupervised learning, weakly supervised learning, and domain adaptation and highlight its applications in computer vision and natural language processing.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4774-4791"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859745","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}

引用次数: 0

Class-Imbalanced Learning via Multitask Multiobjective Feature Selection 基于多任务多目标特征选择的类不平衡学习

IEEE transactions on artificial intelligence Pub Date : 2026-09-01 Epub Date: 2026-03-13 DOI: 10.1109/TAI.2026.3673680

Ruwang Jiao;Naoki Masuyama;Yusuke Nojima

{"title":"Class-Imbalanced Learning via Multitask Multiobjective Feature Selection","authors":"Ruwang Jiao;Naoki Masuyama;Yusuke Nojima","doi":"10.1109/TAI.2026.3673680","DOIUrl":"https://doi.org/10.1109/TAI.2026.3673680","url":null,"abstract":"Class-imbalanced and high-dimensional data pose significant challenges in machine learning, often leading to learning models that underperform on minority classes and struggle with the curse of dimensionality, respectively, and these two issues are inherently interconnected. This article simultaneously addresses both class imbalance and high dimensionality through a multitask multiobjective feature selection method. To be specific, the proposed method establishes two related yet distinct feature selection problem formulations for a given task, each focusing on different aspects of overall and minority classification performance. By selecting important features from the original imbalanced dataset, the proposed method avoids the need to artificially equalize class distribution during training. In addition, the cross-task genetic transfer operator enables the reuse of valuable feature subsets evolved from separate searches, thereby improving both minority class accuracy and overall model performance. Experimental results on 17 class-imbalanced datasets show that the proposed method achieves better results in terms of overall classification performance and minority class accuracy than the compared state-of-the-art methods, particularly on the high-dimensional datasets.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4918-4932"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859718","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}

引用次数: 0

FedDOT: Defending Federated Learning Against Overwhelming Targeted Attacks FedDOT：保护联邦学习免受压倒性的目标攻击

IEEE transactions on artificial intelligence Pub Date : 2026-09-01 Epub Date: 2026-03-23 DOI: 10.1109/TAI.2026.3676747

Priyesh Ranjan;Ashish Gupta;Federico Corò;Sajal K. Das

{"title":"FedDOT: Defending Federated Learning Against Overwhelming Targeted Attacks","authors":"Priyesh Ranjan;Ashish Gupta;Federico Corò;Sajal K. Das","doi":"10.1109/TAI.2026.3676747","DOIUrl":"https://doi.org/10.1109/TAI.2026.3676747","url":null,"abstract":"Federated learning (FL), which facilitates collaborative model training and protects users’ privacy, has drawn great interest from the research community. With FL, participants train their models on local data and submit the corresponding updates for aggregation to a server. While concealing the identities of the participants, FL may attract adversaries in order to hamper the underlying model. In this article, we propose an FL framework, FedDOT, to defend against adversaries performing <italic>targeted attacks. FedDOT incorporates two powerful defense algorithms, maximum spanning tree-based attacker detection (MST-AD) and densest graph-based attacker detection (density-AD), which leverage correlation between weight updates and graph theory concepts, maximum spanning tree, and densest graph. With a goal to withstand an overwhelming number of attackers, our algorithms provide strong solutions to aid an FL server, even in overwhelming scenarios where adversaries constitute more than half of the participants. Along with theoretical bounds in correlation space, a rigorous experimental analysis using image classification datasets is carried out to validate the robustness of the FedDOT framework in non-IID settings, which ascertains the superiority of the models against the state-of-the-art methods using a variety of metrics evaluating the accuracy and attack detection rate. With an attack success rate of <inline-formula><tex-math>$ltmathbf{10%}$</tex-math></inline-formula> for targeted attacks like single-label flipping, multilabel flipping, and backdoor, FedDOT successfully defends against overwhelming adversaries with a marginal accuracy drop of less than <inline-formula><tex-math>$mathbf{2%}$</tex-math></inline-formula>.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5018-5032"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859726","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}

引用次数: 0

Feature Expansion With Semi-Dynamic Feature Sets (SDFS) and LLM-Powered Explainability: A Novel Approach to Discovering Missing Features 半动态特征集（SDFS）和llm驱动的可解释性特征扩展：一种发现缺失特征的新方法

IEEE transactions on artificial intelligence Pub Date : 2026-09-01 Epub Date: 2026-04-07 DOI: 10.1109/TAI.2026.3681539

Elaheh Hosseini;Soodeh Nikan

{"title":"Feature Expansion With Semi-Dynamic Feature Sets (SDFS) and LLM-Powered Explainability: A Novel Approach to Discovering Missing Features","authors":"Elaheh Hosseini;Soodeh Nikan","doi":"10.1109/TAI.2026.3681539","DOIUrl":"https://doi.org/10.1109/TAI.2026.3681539","url":null,"abstract":"Traditional feature selection methods focus on reducing dimensionality, often at the cost of discarding valuable but hidden information. In contrast, this article introduces a feature expansion method, a novel approach that automatically identifies and reconstructs missing features, meaning attributes overlooked or undiscovered during initial feature extraction, enhancing predictive performance without requiring additional data collection. We propose the semi-dynamic feature set (SDFS) network, a unique machine learning framework that learns both feature values and model parameters, thereby improving robustness in complex datasets. To enhance interpretability, we integrate large language models (LLMs) to assign meaningful labels to the newly discovered features, transforming them from black-box variables into actionable insights. Our proposed feature expansion method consistently improved classification and regression performance across multiple datasets. In particular, the accuracy increased from 89% to 95% for Parkinson’s disease detection and from 82% to 85% for wine quality prediction, and reduced the mean squared error (MAE) in air quality prediction from 91.2 to 85.04, demonstrating its effectiveness in discovering missing features. By combining our feature expansion with LLM-driven explainability, this method not only enhances model’s performance, but also unveils hidden data patterns in critical domains such as healthcare, finance, and scientific research. The full implementation is available on GitHub [GitHub link], and the model can be installed via PyPI under the package “sdfs” [PyPI link].","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5224-5238"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859702","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}

引用次数: 0

Stereo Image Coding for Machines With Joint Visual Feature Compression 基于联合视觉特征压缩的机器立体图像编码

IEEE transactions on artificial intelligence Pub Date : 2026-09-01 Epub Date: 2026-03-09 DOI: 10.1109/TAI.2026.3671761

Dengchao Jin;Jianjun Lei;Bo Peng;Zhaoqing Pan;Bo Zhao;Nam Ling;Qingming Huang

引用次数: 0

Utilizing Everyday Textures as Concept-Based Explanations for Deep Learning Model Prediction of Glioma Biomarker Status 利用日常纹理作为基于概念的深度学习模型预测胶质瘤生物标志物状态的解释

IEEE transactions on artificial intelligence Pub Date : 2026-09-01 Epub Date: 2026-04-07 DOI: 10.1109/TAI.2026.3681537

Rayyan Akhand;Farzad Alizadeh;Thanh Binh Nguyen;Nick D. James;Sreeraman Rajan;Rebecca E. Thornhill

{"title":"Utilizing Everyday Textures as Concept-Based Explanations for Deep Learning Model Prediction of Glioma Biomarker Status","authors":"Rayyan Akhand;Farzad Alizadeh;Thanh Binh Nguyen;Nick D. James;Sreeraman Rajan;Rebecca E. Thornhill","doi":"10.1109/TAI.2026.3681537","DOIUrl":"https://doi.org/10.1109/TAI.2026.3681537","url":null,"abstract":"<italic>Objective: This study aims to bridge the gap between artificial intelligence (AI) model prediction and human comprehension of higher level imaging concepts. The specific application involves magnetic resonance imaging (MRI) of gliomas, focusing on AI classification of isocitrate dehydrogenase (IDH) biomarker status, a key indicator of glioma aggressiveness. <italic>Methods: Four convolutional neural network (CNN) models, based on the ResNet50 architecture, were trained using MRI of gliomas to differentiate between gliomas with “wildtype” and “mutated” forms of the IDH biomarker. The models were evaluated using fivefold cross-validation. In addition, the testing concept activation vectors (TCAV) framework was employed to assess the impact of “everyday” image texture concepts on model decision-making. Sensitivity scores ranging from 0 to 1 were calculated for 47 concepts across three layers of each model for both forms of IDH. <italic>Results: The analysis revealed varying levels of sensitivity to different image texture concepts when classifying IDH forms. These sensitivity scores indicate the extent to which the model is influenced by these concepts when making decisions about the IDH type. <italic>Conclusion: This study provides a preliminary evaluation of how concept-based explanations can elucidate the internal workings of closed-box neural networks, particularly in a medical context. <italic>Significance: The findings offer valuable insights into how concept-based explanations can improve understanding and trust in AI models, which is crucial for deploying AI models in clinical settings.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5215-5223"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859738","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}

引用次数: 0

An End-to-End Dual-Attention Based Target Speaker Extraction Network for Explicit Optimization of Speaker Confusion 基于端到端双注意的目标说话人提取网络，用于显式优化说话人混淆

IEEE transactions on artificial intelligence Pub Date : 2026-09-01 Epub Date: 2026-03-30 DOI: 10.1109/TAI.2026.3678540

Hongkai Zhao;Zhelong Wang;Yundong Han;Xiang Li;Hongyu Zhao;Sen Qiu

{"title":"An End-to-End Dual-Attention Based Target Speaker Extraction Network for Explicit Optimization of Speaker Confusion","authors":"Hongkai Zhao;Zhelong Wang;Yundong Han;Xiang Li;Hongyu Zhao;Sen Qiu","doi":"10.1109/TAI.2026.3678540","DOIUrl":"https://doi.org/10.1109/TAI.2026.3678540","url":null,"abstract":"The task of target speaker extraction (TSE) involves isolating a specific speaker’s voice from a mixture of multiple audio sources. Most existing TSE methods primarily optimize the reconstruction quality of the extracted waveform while overlooking essential cues contained in the reference audio. Moreover, traditional TSE networks lack an interactive learning mechanism that jointly models mixed and reference embeddings, often leading to speaker confusion, particularly when there is a notable mismatch between the reference and target audio. To overcome these challenges, we propose a novel end-to-end dual-attention-based target speaker extraction (DATSE) network that explicitly addresses speaker confusion. DATSE employs an attention-weighted multiscale speech encoder that adaptively emphasizes features at different temporal resolutions. In addition, it introduces an attention-based speaker aggregation module within the auxiliary speaker recognition network, which effectively integrates speaker-specific information and leverages mixed-audio cues to refine target speaker embeddings. Experimental results on the WSJ0-2mix-extr dataset show that, the proposed method not only achieves competitive performance across multiple metrics but also significantly reduces speaker confusion under various disturbances such as noise and reverberation, thereby effectively lowering the overall confusion rate.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5105-5116"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859744","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}

引用次数: 0

TrustAI: An Expert-Based Methodology for Evaluating and Certifying the Trustworthiness of Artificial Intelligence Systems TrustAI：一种基于专家的人工智能系统可信度评估与认证方法

IEEE transactions on artificial intelligence Pub Date : 2026-09-01 Epub Date: 2026-03-23 DOI: 10.1109/TAI.2026.3677323

Miltiadis Siavvas;Dimitrios Tsoukalas;Dionusios Kehagias;Dimitrios Tzovaras

{"title":"TrustAI: An Expert-Based Methodology for Evaluating and Certifying the Trustworthiness of Artificial Intelligence Systems","authors":"Miltiadis Siavvas;Dimitrios Tsoukalas;Dionusios Kehagias;Dimitrios Tzovaras","doi":"10.1109/TAI.2026.3677323","DOIUrl":"https://doi.org/10.1109/TAI.2026.3677323","url":null,"abstract":"Despite the acknowledged importance of ensuring the trustworthiness of artificial intelligence (AI) systems, there is currently a lack of well-established methods and techniques for evaluating and certifying their trust. To address this issue, we propose TrustAI, a novel expert-based methodology for evaluating and certifying the trustworthiness of AI systems. Our approach is based on a set of high-level requirements, i.e., the pillars of trust, which are decomposed into a set of low-level criteria expressed in the form of questions that are assessed by AI experts. It aggregates the expert judgments through a hierarchical model to provide a quantitative expression of the level of trust (LoT) of a given AI system and generates a certificate in case that the evaluation results are satisfactory. Fuzzy logic is utilized for modeling the uncertainty of expert judgments and for defining the parameters of the model. The proposed methodology is highly configurable, enabling the generation of custom evaluation models tailored to specific application domains and regulatory frameworks. It is illustrated through a use case on two actual AI systems, evaluated using a model derived via the methodology based on a recently issued regulatory framework.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"5278-5286"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859733","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}

引用次数: 0

Exploiting GANs Against IDSs: A Systematic Review, Meta-Analysis, and Case Study Evaluation 利用gan对抗入侵防御系统：系统回顾、元分析及个案研究评估

IEEE transactions on artificial intelligence Pub Date : 2026-09-01 Epub Date: 2026-03-23 DOI: 10.1109/TAI.2026.3677234

Yasmeen Alslman;Mouhammd Alkasassbeh;Mohammad J. Abdel-Rahman

{"title":"Exploiting GANs Against IDSs: A Systematic Review, Meta-Analysis, and Case Study Evaluation","authors":"Yasmeen Alslman;Mouhammd Alkasassbeh;Mohammad J. Abdel-Rahman","doi":"10.1109/TAI.2026.3677234","DOIUrl":"https://doi.org/10.1109/TAI.2026.3677234","url":null,"abstract":"With the rapid improvement of generative adversarial networks (GANs) in the landscape of data augmentation and generation, researchers have started to explore the use of GANs in other fields, such as adversarial attack generation. Using GANs has improved the field of adversarial attacks, particularly against intrusion detection systems (IDSs). GAN-based adversarial attacks are considered sophisticated attacks that can significantly compromise the security of IDSs. Addressing GAN-based adversarial challenges can ensure the effectiveness and severity of such attacks, which, in turn, lead to the development of strong defense strategies. Few surveys have covered the effects and challenges of adversarial attacks against IDSs, with a focus on gradient-based adversarial attacks. Recently, many research articles have examined GAN-based adversarial attacks. This article provides a systematic and comprehensive review of recent research on GAN-based adversarial attacks targeting IDSs, complemented by a meta-analysis guided by critical review questions. Through case studies, we empirically demonstrate the vulnerabilities of binary and multiclass IDS models under GAN-based attacks, examining the influence of feature selection. We are the first to employ advanced GAN variants to craft novel adversarial attacks. Using Fréchet inception distance and Hellinger distance, we assess IDS performance and the realism of generated samples, revealing key system weaknesses. The article concludes by summarizing the current state of knowledge, outlining major challenges, proposing future research directions, and identifying the most severe GAN architecture and the one that generates the most realistic adversarial samples.","PeriodicalId":73305,"journal":{"name":"IEEE transactions on artificial intelligence","volume":"7 9","pages":"4751-4773"},"PeriodicalIF":0.0,"publicationDate":"2026-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"148859742","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}

引用次数: 0

Reinforcement Learning-Based Active Fault-Tolerant Control of Multi-Agent Systems 基于强化学习的多智能体系统主动容错控制

IEEE transactions on artificial intelligence Pub Date : 2026-09-01 Epub Date: 2026-03-30 DOI: 10.1109/TAI.2026.3678538

Sikai Shen;Xiaoyang Liu;Wenwu Yu

引用次数: 0