IEEE transactions on neural networks and learning systems最新文献

{"title":"Generative Learning Imaging Framework for Millimeter Wave Synthetic Aperture Radar.","authors":"Mou Wang, Yifei Hu, Hao Zhang, Xiang Cai, Jiahui Chen, Shunjun Wei, Jun Shi, Guolong Cui, Lingjiang Kong, Yongxin Guo","doi":"10.1109/TNNLS.2026.3686777","DOIUrl":"https://doi.org/10.1109/TNNLS.2026.3686777","url":null,"abstract":"<p><p>Accurately reconstructing synthetic aperture radar (SAR) images from incomplete echo measurements is the fundamental problem in simplifying SAR systems and reducing sensing costs. In solving this problem, the recently emerged deep learning SAR imaging concept shows high potential in offering both high accuracy and efficiency. However, such approaches face challenges in network topology designing, weights initialization, and dataset construction, which hinder their practical deployments. To address these problems, we propose a generative learning imaging framework, namely, generative learning imaging framework (GLIm), for millimeter wave (mmWave) SAR in tasks of sparsely sensing scenarios. In our scheme, the targeting images are generated from a specifically designed SAR image generator by mapping low-dimensional random noise to the scattering domain. In addition, the SAR image generator is trained based on a compound loss function where the incomplete echo measurements serve as the supervised signal. Unlike traditional algorithms, the algorithm proposed in this article solves the optimal solution purely through numerical propagation by online learning, avoiding the theoretical derivation of the iterative optimization process. Finally, experiments are carried out on both simulated and real-measured data, and both numerical and visual results demonstrate the viability of the proposed GLIm in the tasks of reconstructing mmWave SAR images from sparsely sampled echo measurements.</p>","PeriodicalId":13303,"journal":{"name":"IEEE transactions on neural networks and learning systems","volume":"PP ","pages":""},"PeriodicalIF":8.9,"publicationDate":"2026-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147856268","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Graph Rotation Network: Equivariant Graph Neural Network for Efficient Inverse Design in 4-D Printing.","authors":"Yue Wang, Mingjun Tang, Shen Gao, Zhilin Chen, Na Liu, Yuzhao Zhang, Xiaodong Yue, Tao Yue, Yuxuan Yu, Chao Wang, Shaorong Xie, Wen Jung Li","doi":"10.1109/TNNLS.2026.3679129","DOIUrl":"https://doi.org/10.1109/TNNLS.2026.3679129","url":null,"abstract":"<p><p>Four-dimensional printed structures transform shape under external stimuli, and designing these structures to achieve desired deformation outcomes is crucial for 4-D printing technology. Current machine-learning (ML)-assisted methods can design 4-D-printed structures rapidly and accurately, but often overlook the E(3) equivariance required for the model, leading to inconsistent prediction structure between different coordinate representations of the same structure. Consequently, many ML-assisted design approaches rely on data augmentation techniques such as rotation to improve the model's generalization across different coordinate representations, but the method through data augmentation is inefficient. We observe that equivariant graph neural networks (GNNs), due to their equivariant properties, can address the inefficiencies associated with the aforementioned training augmentation. Therefore, we propose utilizing an equivariant GNN to improve these efficiencies. Nonetheless, existing equivariant GNNs struggle to balance computational efficiency with expressive power, which will impact their effectiveness in 4-D printing structure design. In response, we introduce a novel equivariant GNN, the graph rotation network (GRN), which approximates arbitrary E(3) equivariant functions by constructing a 3-D equivariant basis through rotations. This approach enhances the model's performance with trainable rotational angles. We theoretically analyze the E(3) equivariance of our model as well as its computational complexity and empirically demonstrate the improved training efficiency introduced by equivariance. Finally, we apply GRN to both forward prediction and inverse design of 4-D-printed meshes. Our equivariant GNN significantly outperforms nonequivariant GNNs and other equivariant GNNs in forward prediction tasks. For inverse design, our method achieves an accuracy of 98.8% (relative to the mesh size), providing an effective tool for the design of 4-D-printed structures.</p>","PeriodicalId":13303,"journal":{"name":"IEEE transactions on neural networks and learning systems","volume":"PP ","pages":""},"PeriodicalIF":8.9,"publicationDate":"2026-05-08","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147856230","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Explainable Multihop Social Link Prediction Based on Temporal Logical Rules in Dynamic Social Networks.","authors":"Wei Jia, Ruizhe Ma, Li Yan, Weinan Niu, Zongmin Ma","doi":"10.1109/TNNLS.2026.3680219","DOIUrl":"https://doi.org/10.1109/TNNLS.2026.3680219","url":null,"abstract":"<p><p>The social link prediction poses a fundamental challenge in social network analysis, aiming to forecast missing interactions among users. Given the dynamic evolution mechanism of social networks, prevailing efforts have introduced embedding-based approaches to address temporal link prediction in dynamic social networks. However, these approaches often struggle to handle explainability, multirelations, and multihop relation prediction simultaneously. To overcome these limitations, we present an innovative multihop temporal social link prediction model based on temporal logic embedding (TLE), which leverages temporal knowledge graphs and logic rules. First, inspired by the temporal knowledge graph, we construct temporal social knowledge graphs (TSKGs) to model dynamic social networks. Then, we incorporate the orthogonal transformation matrix into the graph neural networks (GNNs), thereby facilitating the learning of time-aware relation representations. Furthermore, we define temporal social random walks from the TSKG to generate temporal social rules. Subsequently, TLE employs time-aware relation embedding to calculate the confidence associated with each rule. Finally, TLE combines the confidence score and time difference to obtain the final link prediction, providing explainability for multihop link predictions. The experiments carried out on four datasets indicate the superiority of TLE in the social link prediction within dynamic social networks.</p>","PeriodicalId":13303,"journal":{"name":"IEEE transactions on neural networks and learning systems","volume":"PP ","pages":""},"PeriodicalIF":8.9,"publicationDate":"2026-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147837359","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Input-to-State Safety for Reinforcement Learning.","authors":"Mayank Shekhar Jha, Satya Marthi, Kyriakos G Vamvoudakis, Soha Kanso, Didier Theilliol","doi":"10.1109/TNNLS.2026.3688045","DOIUrl":"https://doi.org/10.1109/TNNLS.2026.3688045","url":null,"abstract":"<p><p>In this article, we present a novel off-policy, safe reinforcement learning (RL) approach for nonlinear dynamical systems under input saturation that guarantees safe initialization, safe exploration, as well as safe learning of optimal control laws. First, to encourage preferable exploration near safety boundaries, important for integrating system behavior near the safety limits, we formulate a safe exploration approach as a robust control problem by considering an enlarged safe set based on input-to-state safe control barrier functions (ISSf-CBFs). These constraints are then incorporated into a quadratic programming (QP) optimization. We propose a novel $epsilon $ -tuning law that adaptively enforces stricter safety constraints near the boundaries of the safe set and relaxes constraints deeper within the safe set, encouraging safety boundary-proximal exploration while maintaining forward invariance of the safe set. The proposed $epsilon $ -tuning law safely accommodates aggressive, high-magnitude exploration noise, enabling efficient state-space exploration without compromising safety. Next, safe learning under saturation limits is guaranteed through a safety-aware cost function. We establish safety, optimality, and stability properties (novel) in a mathematically rigorous manner. Furthermore, the safe RL problem is solved in an off-policy manner, and neural networks are used to approximate the value function and the control policy. To that end, we establish a novel off-policy equation under input saturation. Finally, simulations demonstrate the efficacy of the proposed framework.</p>","PeriodicalId":13303,"journal":{"name":"IEEE transactions on neural networks and learning systems","volume":"PP ","pages":""},"PeriodicalIF":8.9,"publicationDate":"2026-05-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147837432","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"GBFRS: Robust Fuzzy Rough Sets via Granular Ball Computing.","authors":"Xiaoyu Lian, Shuyin Xia, Binbin Sang, Guoyin Wang, Xinbo Gao","doi":"10.1109/TNNLS.2026.3681370","DOIUrl":"https://doi.org/10.1109/TNNLS.2026.3681370","url":null,"abstract":"<p><p>Fuzzy rough set (FRS) theory is effective for handling uncertainty and vagueness in complex datasets. However, most existing models rely on fine-grained pointwise analysis, resulting in poor robustness to noise. This article integrates granular ball computing into FRSs by replacing individual data points with granular balls of varying sizes. A granular ball FRS (GBFRS) framework is proposed and applied to feature selection for the first time. Each granular ball is labeled by the majority class of its internal samples, reducing the impact of noisy instances and improving noise tolerance. Within this framework, a weighted fuzzy dependency function is redefined based on fixed https://github.com/lianxiaoyu724/GBFRS structures, where the weight is determined by the proportion of samples within each granular ball. Larger balls have higher fuzzy dependency values and thus receive greater weights, enabling a more stable evaluation of attribute importance. The theoretical foundations, including properties of lower and upper approximations and the convergence of dependency, are formally established. The experimental results on multiple UCI datasets demonstrate that GBFRS outperforms existing methods in classification accuracy. The source codes and datasets are both available on the public link: https://github.com/lianxiaoyu724/GBFRS.</p>","PeriodicalId":13303,"journal":{"name":"IEEE transactions on neural networks and learning systems","volume":"PP ","pages":""},"PeriodicalIF":8.9,"publicationDate":"2026-05-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147837391","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Topology-Preserving Deep Hashing for Ultrafast Drone-Dominated Object Detection.","authors":"Luming Zhang, Guifeng Wang, Zhiming Wang, Ling Shao","doi":"10.1109/TNNLS.2026.3686846","DOIUrl":"https://doi.org/10.1109/TNNLS.2026.3686846","url":null,"abstract":"<p><p>Drone (or unmanned aerial vehicle) has been extensively applied in many modern artificial intelligence systems in the past decade. In this work, we propose a novel deep hashing framework that can detect objects from drone-captured pictures extremely fast. Our method can intrinsically and flexibly encode various topological structures from each target object, based on which multiscale objects can be discovered in a view- and altitude-invariant way. Moreover, by leveraging $l_{F}$ and $l_{1}$ norms collaboratively, the calculated hash codes are robust to low-quality drone pictures and possibly contaminated semantic labels. More specifically, for each drone picture, we extract visually/semantically salient object parts inside it. To characterize their topological structure, we construct a graphlet by linking the spatially adjacent object patches into a small graph. Subsequently, a binary matrix factorization (MF) is designed to hierarchically exploit the semantics of these graphlets, wherein three attributes: 1) deep binary hash codes learning; 2) contaminated pictures/labels denoising; and 3) adaptive data graph updating are seamlessly incorporated. Accordingly, a manifold-regularized feature selector is adopted to further obtain more discriminative deep hash codes. Finally, the selected hash codes corresponding to graphlets within each drone photograph are utilized for ranking-based object discovery. Comprehensive experiments on the DAC-SDC, MOHR, and our self-compiled dataset have demonstrated the competitive speed and accuracy of our method.</p>","PeriodicalId":13303,"journal":{"name":"IEEE transactions on neural networks and learning systems","volume":"PP ","pages":""},"PeriodicalIF":8.9,"publicationDate":"2026-05-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147837362","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"MoFTSS: Motion Generation With Frequency and Text State Space Models.","authors":"Chengjian Li, Xiangbo Shu, Qiongjie Cui, Haifeng Xia, Yazhou Yao, Jinhui Tang","doi":"10.1109/TNNLS.2026.3683909","DOIUrl":"https://doi.org/10.1109/TNNLS.2026.3683909","url":null,"abstract":"<p><p>Text-driven diffusion models have achieved remarkable performance in human motion generation. However, these generative works struggle to generate high-quality motion consistent with textual descriptions. The primary reasons are: 1) insufficient fine-grained motion modeling due to the motion representations being difficult to distinguish in latent diffusion; and 2) inconsistencies between motions and textual descriptions due to misalignment in the multimodal space. To overcome these limitations, this work proposes the Motion generation with Frequency and Text State Space models (MoFTSS) including two main modules: frequency state space model (FreqSSM) and text state space model (TextSSM). Specifically, FreqSSM derives fine-grained representations by decomposing sequences into low-frequency and high-frequency components. This allows it to guide the generation of static poses (e.g., sitting, lying) and fine-grained motions (e.g, transitions, stumbling). For consistency between text and motion, TextSSM treats text features as a semantic modulation term within the SSM, enabling dynamic filtering of motion features consistent with textual semantics. Extensive experiments suggest that our MoFTSS achieves superior performance on the text-to-motion generation task. Notably, it attains the lowest FID of 0.181 on the HumanML3D dataset, significantly lower than the 0.421 achieved by MLD.</p>","PeriodicalId":13303,"journal":{"name":"IEEE transactions on neural networks and learning systems","volume":"PP ","pages":""},"PeriodicalIF":8.9,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147814021","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Deep Reinforcement Learning-Based Optimization of Identical-Dual-Band Filters.","authors":"Ehsan Adibnia","doi":"10.1109/TNNLS.2026.3684954","DOIUrl":"https://doi.org/10.1109/TNNLS.2026.3684954","url":null,"abstract":"<p><p>Designing identical dual-band optical filters remains a complex optimization challenge in photonics and optical communication systems. Conventional methods, which rely on iterative electromagnetic simulations or analytical approximations, often suffer from limited generalizability and high computational costs. In this work, we propose a deep reinforcement learning (RL) framework for the autonomous optimization of identical dual-band fiber Bragg grating (FBG) filters. A policy network based on a three-layer fully connected neural architecture is trained using a proximal policy optimization algorithm to minimize the full width at half maximum (FWHM) of both transmission bands while maintaining spectral symmetry and identical channel characteristics. The deep RL-based design achieves a 43% reduction in FWHM and a 49% reduction in grating length compared to baseline designs, without sacrificing reflectivity or channel uniformity. This study demonstrates the feasibility and effectiveness of deep RL as a powerful optimization tool for complex photonic systems, providing a scalable and data-efficient pathway toward next-generation optical device design.</p>","PeriodicalId":13303,"journal":{"name":"IEEE transactions on neural networks and learning systems","volume":"PP ","pages":""},"PeriodicalIF":8.9,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147814588","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"XAI-Exit: Interpretability-Driven Dynamic Early Exits for Efficient and Transparent DNN Inference.","authors":"Haseena Rahmath P, Ajith Abraham, Kuldeep Chaurasia","doi":"10.1109/TNNLS.2026.3685408","DOIUrl":"https://doi.org/10.1109/TNNLS.2026.3685408","url":null,"abstract":"<p><p>Deep neural networks (DNNs) excel across domains but face challenges in resource-constrained and critical settings due to high computational cost and limited transparency. Early exit DNNs reduce overhead via intermediate predictions; yet, most approaches neglect interpretability, vital for trust in AI systems. This article presents XAI-Exit, an early exit framework that jointly optimizes efficiency and transparency. At its core, ExitDecisionNet (EDN)-a lightweight RNN trained with a curriculum strategy on confidence, interpretability, and stability metrics-dynamically predicts the optimal exit, while a skip mechanism minimizes redundant computation. To ensure transparency, exit attribution maps (EAMs) aggregate feature attributions across exits, revealing the decision trajectory and are complemented by standard XAI methods (integrated gradients (IGs), SmoothGrad, Grad-CAM++, and LRP). Experiments on MobileNetV3, ResNet18, and MSDNet with CIFAR-10, CIFAR-100, and ImageNet show that XAI-Exit improves efficiency without sacrificing accuracy, while uniquely ensuring interpretable exit decisions suitable for real-world deployment.</p>","PeriodicalId":13303,"journal":{"name":"IEEE transactions on neural networks and learning systems","volume":"PP ","pages":""},"PeriodicalIF":8.9,"publicationDate":"2026-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147814072","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"MID: A Self-Supervised Multimodal Iterative Denoising Framework.","authors":"Chang Nie, Tianchen Deng, Zhe Liu, Hesheng Wang","doi":"10.1109/TNNLS.2026.3683544","DOIUrl":"https://doi.org/10.1109/TNNLS.2026.3683544","url":null,"abstract":"<p><p>Denoising is important in many vision, medical, and biological applications, yet real observations are often corrupted by complex nonlinear noise and clean targets are often unavailable. We present MID, a self-supervised iterative denoising framework across data modalities. MID treats an observation as an intermediate state along a controllable corruption process and learns from noisy data only through two networks: a step predictor that estimates the current corruption stage and a residual predictor that estimates the effective residual increment to be removed at that stage. For nonlinear corruption, MID uses a first-order local approximation to enable iterative restoration in a locally linear regime. The same formulation can be instantiated with modality-specific backbones for images, signals, point sets, and sequences. Experiments on diverse tasks in computer vision, biomedicine, and bioinformatics show that MID is robust, broadly applicable, and competitive with recent baselines.</p>","PeriodicalId":13303,"journal":{"name":"IEEE transactions on neural networks and learning systems","volume":"PP ","pages":""},"PeriodicalIF":8.9,"publicationDate":"2026-04-30","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147813843","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":1,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}