Fine-Grained Domain Generalization (FGDG) presents greater challenges than conventional domain generalization due to the subtle inter-class differences and relatively pronounced intra-class variations inherent in fine-grained recognition tasks. Under domain shifts, the model becomes overly sensitive to fine-grained cues, leading to the suppression of critical features and a significant drop in performance. Cognitive studies suggest that humans classify objects by leveraging both common and specific attributes, enabling accurate differentiation between fine-grained categories. However, current deep learning models have yet to incorporate this mechanism effectively. Inspired by this mechanism, we propose Concept-Feature Structuralized Generalization (CFSG). This model explicitly disentangles both the concept and feature spaces into three structured components: common, specific, and confounding segments. To mitigate the adverse effects of varying degrees of distribution shift, we introduce an adaptive mechanism that dynamically adjusts the proportions of common, specific, and confounding components. In the final prediction, explicit weights are assigned to each pair of components. Extensive experiments on three single-source benchmark datasets demonstrate that CFSG achieves an average performance improvement of 9.87% over baseline models and outperforms existing state-of-the-art methods by an average of 3.08%. Additionally, explainability analysis validates that CFSG effectively integrates multi-granularity structured knowledge and confirms that feature structuralization facilitates the emergence of concept structuralization.

Class Incremental Learning (CIL) aims to enable models to continually learn new classes while retaining previously learned knowledge. The principal challenge in CIL is catastrophic forgetting, which prior approaches typically address by distilling knowledge from previous model. However, such way is often limited to pairwise alignment, failing to preserve the underlying global manifold structure of feature space—ultimately resulting in semantic drift over time. To capture multi-scale structural patterns in the feature space, we propose a topology-aware distillation framework that leverages persistent homology. Specifically, by enforcing topological alignment across incremental stages, our method ensures structure-consistent knowledge transfer and robust preservation of old classes. Furthermore, we still devise a dual-branch architecture with an inverse sampling and dynamic reweighting mechanism that addresses the inherent data imbalance in standard replay-based frameworks. These innovations coalesce into TaKP (Topology-aware Knowledge Preservation), a unified framework designed to enhance knowledge preservation in CIL. Extensive experiments demonstrate that TaKP achieves state-of-the-art performance on multiple benchmarks, significantly improving old-class preservation and average accuracy.

To address the issues of sparse user interaction data and insufficient mining of knowledge features from the perspective of single users in online education platforms, a group-knowledge enhanced personalized educational resource recommendation model (GEPR) is proposed.First, a user-resource interaction graph is constructed based on multi-user interaction sequences, and user groups are iteratively partitioned through behavioral features.Next, the complex associations between resources and knowledge points are mined from text information, and a semantic information constraint term is designed to optimize the knowledge representation learning of resource entities, enhancing resource representation in both structural and textual dimensions.Then, a cross-attention mechanism with group knowledge constraints is designed to establish group-individual knowledge propagation channels, achieving the fusion of group and individual knowledge features through adaptive adjustment of attention weights.Finally, group knowledge information is integrated into the state representation and reward function of reinforcement learning (RL)to optimize the recommendation strategy. Comparative experiments with nine recommendation methods on two online education datasets show that GEPR can fully combine the semantic information of resources and group knowledge information, capture users’ personalized needs, and improve recommendation accuracy.

Unsupervised deep clustering plays a crucial role in analyzing single-cell RNA sequencing data (scRNA-seq) as it helps to identify potential cell types. However, most existing clustering methods face challenges in effectively fusing common information between feature and topological structure information, and they may not perform well on the sparse data, which are common in single-cell analysis. To address these challenges, we propose a novel Graphical Deep Clustering with Fused Common Information method for scRNA-seq data, named scGDCF. This method can accurately segregate different cell types even in large and sparse scRNA-seq datasets. In scGDCF, we first introduce a sparse feature representation method that utilizes an adversarial loss function to address the sparsity problem in scRNA-seq data and improve the performance of the discriminator. Next, we design a mutual information extracting operator to deeply mine and fuse the common information from feature and topological structure data, thereby improving the clustering performance. Furthermore, we incorporate the varying degrees of contribution from different neighbor nodes and information sources. To handle this, we promote a dual adaptive attention mechanism that operates at both global and local levels. Finally, experiments on seven real-world datasets and two simulated datasets show scGDCF outperforms 17 state-of-the-art methods. We further extend the clustering results for visualization, analysis of gene differential expression and enrichment, showing scGDCF provides novel insights into cell developmental lineages and preserved inter-cluster distances.

Cell clustering is crucial in single-cell multi-omics research for identifying distinct cellular populations. Although there has been progress in integrating multi-omics data for clustering, combining more than two types of omics data remains challenging due to the diversity and heterogeneity of these datasets. Traditional approaches typically use heterogeneous graphs that integrate only two types of omics data, constructing graphs with genes and cells as nodes and a single type of edge representing their relationships. However, this method has limitations as it overlooks cell-cell interactions and struggles to capture complex cellular dynamics. Additionally, the graph structure must be redesigned whenever new omics data are introduced, limiting the scalability of these models. To address these issues, we introduce MOH, a novel single-cell clustering algorithm based on a multilayer multi-omics heterogeneous graph. MOH integrates three key single-cell omics types: scRNA-seq, scATAC-seq, and spatial transcriptomics. It constructs a multilayer heterogeneous graph to simultaneously extract and enhance representations from all three omics layers, incorporating both intra-layer and inter-layer edges to capture association and similarity relationships. This enriched representation leads to an accurate clustering results. Extensive experiments show that MOH outperforms six state-of-the-art methods on unsupervised clustering metrics, offering a precise and comprehensive analysis with consistent improvements across all evaluation criteria. Moreover, downstream analyses validate the results, revealing novel biological insights into immune disorder complications in cancer, cancer drug repurposing, and new signaling pathways, which merit further investigation and validation.

Multimodal image fusion integrates infrared and visible images by leveraging their complementary strengths. However, most existing fusion techniques primarily focus on pixel level integration, often neglecting the preservation of semantic consistency between the source and fused images. To address this limitation, we propose TEDFuse, a Task-Driven Equivariant Consistency Decomposition Network that ensures semantic con sistency within the image space and across high-level semantic tasks. TEDFuse incorporates two key components: first, a robust decomposition framework with equivariant consistency, ensuring that the fused image retains consistent transformation properties under shifts, rotations, and reflections, thereby enhancing local detail preservation and global semantic alignment; In addition, a task-driven fusion framework that integrates a segmentation module, reinforcing semantic feature preservation through a semantic loss function and ensuring consistency in downstream tasks such as segmentation and detection. The proposed method not only preserves the semantic coherence of the fused image but also improves performance in high-level tasks, demonstrating superior capability in multimodal fusion for complex visual applications. Extensive experiments validate the effectiveness of TEDFuse by analyzing feature evolution, examining the relationship between fusion quality and task performance, and discussing calibration strategies for infrared-visible image fusion. The code is available at https://github.com/Claire-cxy/TEDFuse.

Few-Shot Class-Incremental Learning (FSCIL) aims to continuously learn new classes from a limited set of training samples without forgetting knowledge of previously learned classes. Conventional FSCIL methods typically build a robust feature extractor during the base training session with abundant training samples and subsequently freeze this extractor, only fine-tuning the classifier in subsequent incremental phases. However, current strategies primarily focus on preventing catastrophic forgetting, considering only the relationship between novel and base classes, without paying attention to the specific decision spaces of each class. To address this challenge, we propose a plug-and-play Adaptive Decision Boundary Strategy (ADBS), which is compatible with most FSCIL methods. Specifically, we assign a specific decision boundary to each class and adaptively adjust these boundaries during training to optimally refine the decision spaces for the classes in each session. Furthermore, to amplify the distinctiveness between classes, we employ a novel inter-class constraint loss that optimizes the decision boundaries and prototypes for each class. Extensive experiments on three benchmarks, namely CIFAR100, miniImageNet, and CUB200, demonstrate that incorporating our ADBS method with existing FSCIL techniques significantly improves performance, achieving overall state-of-the-art results.

DNNs inevitably over-fits to label noisy sample, resulting in poor generalization. To mitigate the influence of label noise, Co-teaching based methods tend to select potentially clean data as the training samples based on small-loss criterion. However, the small-loss criterion cannot guarantee the purity of clean samples, and blithely ignores the information involved in the noisy samples. To address this issue, we propose a simple yet effective method called SLDR. Specifically, we partition training samples into clean and noisy subsets based on the small-loss criterion to filter the errors generated in each network. Then, we employ supervised regularization to minimize intra-class distances and maximize inter-class distances in clean samples and unsupervised regularization to assess the similarity between instance features in noisy samples. Extensive experiments on mainstream benchmarks, including synthetic noisy datasets (noisy version and long-tailed noisy version), and real-world noisy datasets demonstrate superiority of the proposed method.

In this paper, we focus on the issue of Couple Noisy Labels (CNL) in Visible-Infrared Person Re-identification. CNL which refers to the Noisy Annotations and the Noisy Correspondences. Existing methods have a drawback of wasting samples, as only clean samples selected based on confidence are considered for training. This means that samples with ambiguous predictions do not contribute to the training phase. We propose a robust method dubbed Coupled Noisy with Graph Consistency (CNGC), takes a graph perspective and consists of two components: Node Consistency and Edge Consistency. Node consistency tackles the issue of samples that are discarded due to noisy annotations, while edge consistency addresses the problem of noisy training pairs where both samples are incorrectly labeled. To validate the effectiveness of our method, we conduct extensive experiments on SYSU-MM01 and RegDB datasets. The results demonstrate that CNGC outperforms seven state-of-the-art methods when dealing with Couple Noisy Labels.

It is a challenging task to train robust deep neural networks (DNNs) in the presence of label noise. Though numerous noise robust techniques achieve promising classification performance, little work has theoretically investigated the negative effect of label noise and most of them employ heuristic knowledge and are computationally expensive. In this work, we first conduct a theoretical analysis on label noise from the perspective of gradient back-propagation and find the non-zero gradients derived from noisy labels actually cause performance degradation. Based on this insight, we propose a novel Contrastive Re-Calibration Learning (CRL) framework, which consists of two key components: a probability Re-Calibration Mechanism (RCM) and a Contrastive Calibration Loss (CCL). The RCM introduces a probability re-calibration operation between the softmax function and cross entropy loss by employing a function g, making the gradients derived from noisy labels close to 0 and thus reducing the chance of overfitting to noisy samples. We theoretically infer the conditions of g and propose a set of feasible g functions. The CCL further enhances the model’s ability to distinguish between clean and noisy samples through contrastive learning. Actually, CRL can easily be integrated into other noise-robust methods to further improve the performance of these methods. We verified the limitation effect of noisy labels of CRL by experiments and we conduct extensive experiments on both computer vision tasks (CIFAR), natural language processing tasks (Twitter) and other noisy datasets (Clothing1M). Experimental results show CRL significantly outperforms the state-of-the-art methods.

In this paper, we propose a novel framework named CoCF to address the inconsistency issue in Semi-Supervised Learning (SSL) and Noisy Label Learning (LNL). By integrating SSL methods into the field of LNL, CoCF can gradually and accurately divide clean and noisy data into labeled and unlabeled data, and employ different strategies to process these data separately, thus achieving accurate predictions for clean data and effective utilization of noisy data. Specifically, CoCF adopts the small loss criterion for sample division and adjusts the distribution consistency between labeled and unlabeled data based on Kullback-Leibler divergence. To balance the distribution, CoCF designs a flexible masking mechanism, which includes a dual-branch network structure to train both standard classifier and class-balanced classifier simultaneously, thereby mitigating the negative impact of long-tail datasets on classifier learning. Additionally, CoCF introduces post-hoc logistic adjustment and dynamic logistic adjustment techniques to generate more accurate pseudo labels and further optimize model performance. Experimental results on multiple benchmark and real-world datasets demonstrate that CoCF has significant advantages in handling noisy data, effectively improving the robustness and accuracy of the model.

Single-cell multi-omics clustering confronts noise and heterogeneity barriers. Current multi-view anchor graph approaches, though successful in noise reduction, inadequately model higher order feature interactions. To address this issue, we propose scAGCI, a cell clustering method based on anchor graphs that integrates both scRNA-seq and scATAC-seq data. Our method captures specific and shared anchor graphs representing the properties of omics data in the process of dynamic anchor unification, and mines high-order shared information to complete the omics representation. Subsequently, clustering results are obtained by integrating the specific and shared omics representation. Benchmarking against 13 state-of-the-art methods confirms scAGCI's superior clustering performance and computational efficiency in cell-type identification and subtype resolution. The method preserves biologically meaningful omics patterns, as evidenced by marker gene enrichment and functional analyses, establishing it as a robust tool for elucidating cellular heterogeneity in single-cell multi-omics data.

Fine-grained vehicle classification, which is a key technology within intelligent transportation systems, has been gaining increasing importance with the burgeoning growing number of vehicles. Previous studies have predominantly focused on intricate and distinctive local features. However, in various tasks, it has been proven that global features are of significant importance when they can be effectively integrated with local features in a harmonious manner. So, we consider that a comprehensive consideration of both local and global features is crucial for enhancing classification decisions. Consequently, the paper designs a novel architecture for the task, which combines global and local features to improve classification performance. The architecture consists of two components: the local-feature net and the global-feature net. Specially, for the local feature, we propose an Essential Part Locator module that uses global feature-weighted attention masks to obtain local features, and a Cross-Part Feature Transformer that boosts interactions between local features. Meanwhile, our architecture processes the entire image through an encoder to capture global features and then integrates both global and local features. Experimental results on the Stanford Cars, CompCars, and BoxCars116K datasets demonstrate that the proposed approach surpasses state-of-the-art methods, achieving accuracies of 97.5%, 96.4%, and 92.1%, respectively.

Image retrieval aims to find the most semantically similar images in the database. Existing deep hash-based retrieval algorithms utilize data augmentation strategies thus generating generalized hash codes. However, simple data augmentation only improves the accuracy of hash codes from the perspective of sample diversity, without fully utilizing the inherent characteristics of the images. In this letter, we explore the frequency domain information of images and propose a Frequency Domain Auxiliary Network (FDANet) for deep hash retrieval. To capture frequency domain information that can cope with image transformations, we develop the spectrum enhancement module (SEM) in FDANet. The SEM utilizes Fourier transform techniques to extract the amplitude component that can reflect the low-level statistics of the image. Then, leveraging the extracted amplitude components, the retrieval network enhances its perception of regions undergoing relative changes in the original spatial domain. Experiments on several image retrieval benchmarks demonstrate that our method outperforms other state-of-the-art hash algorithms in terms of performance on the test metrics.

Training noise-robust deep neural networks (DNNs) in label noise scenario is a crucial task. In this paper, we first demonstrates that the DNNs learning with label noise exhibits over-fitting issue on noisy labels because of the DNNs is too confidence in its learning capacity. More significantly, however, it also potentially suffers from under-learning on samples with clean labels. DNNs essentially should pay more attention on the clean samples rather than the noisy samples. Inspired by the sample-weighting strategy, we propose a meta-probability weighting (MPW) algorithm which re-weights the output probability of DNNs to prevent DNNs from over-fitting to label noise and alleviate the under-learning issue on the clean sample. MPW conducts an approximation optimization to adaptively learn the probability weights from data under the supervision of a small clean dataset, and achieves iterative optimization between probability weights and network parameters via meta-learning paradigm. The ablation studies substantiate the effectiveness of MPW to prevent the deep neural networks from overfitting to label noise and improve the learning capacity on clean samples. Furthermore, MPW achieves competitive performance with other state-of-the-art methods on both synthetic and real-world noises.

Ship detection from synthetic aperture radar (SAR) images is a hot topic, but the difficulty in collecting labeled SAR images may hinder the development of deep-learning-based detection methods. Inspired by the idea of domain adaptation, in this article, we propose a hierarchical similarity alignment neural network (HSANet) for ship detection in SAR images, which is a domain adaptive (DA) approach with optical remote sensing images as training samples. The kernel target of HSANet is to mine and align both the global structure and the local instance information between SAR and optical images, where two modules, structural alignment module (SAM) and prototype alignment module (PAM), are designed to, respectively, conduct two hierarchies of alignment process. In general, SAM attempts to extract the global structure similarity which exists in image-level feature representation, while PAM tends to extract the local shape similarity which is instance-level representation. To be specific, SAM is developed by Fourier-based feature alignment, which tries to describe the similar structural relationship between optical and SAR images. Meanwhile, PAM is proposed based on the conjoint confidence analysis where the instance-level ship representations of the source and target domains is aligned. SAM and PAM work together to construct a hierarchical domain adaptation network for SAR ship detection. Experiments on several public datasets may indicate the effectiveness of the proposed method.

Intelligent video surveillance is a vital technique in smart city construction, where detection of surveillance objects is generally achieved by subtracting estimated background from the raw video. Common wisdom of background estimation focuses on introducing meaningful structure or discriminative hypothesis to sparsity-based objectives. However, relaxation optimization, which is always considered a most effective solution, definitely leads to information loss. So, in this article, as to preserve more information, a new nonconvex sparsity model that can be solved directly by explicit solution is proposed for the stationary component of video. The solution, called generalized shrinkage thresholding operator, is designed by integrating the advantages of three common shrinkage operators. Then, for the regularly changing patterns, a purified dictionary learning operation is designed to find self-repeating texture patches. Eventually, foreground objects are detected by combining background subtraction with a spatiotemporal continuity constraint. Besides, built on optimizations of both models, we then show the way to refine the joint estimates using alternative optimization of all the subproblems. Experimental results have shown that, as to foreground detection task, when compared against current state-of-the-art techniques, the proposed model achieves comparable and often superior performance in terms of F-measure scores in most cases.

In copy-move forgery, the illumination and contrast of tampered and genuine regions are highly consistent, which poses a greater challenge in copy-move forgery detection. In this article, an end-to-end neural network is proposed based on adaptive attention and residual refinement network (AR-Net). Specifically, position and channel attention features are fused by the adaptive attention mechanism to fully capture context information and enrich the representation of features. Second, deep matching is adopted to compute the self-correlation between feature maps, and atrous spatial pyramid pooling fuses the scaled correlation maps to generate the coarse mask. Finally, the coarse mask is optimized through the residual refinement module, which retains the structure of object boundaries. Extensive experiments, evaluated on CASIAII, COVERAGE, and CoMoFoD datasets, demonstrate that the AR-Net has superior performance than state-of-the-art algorithms and can locate tampered and corresponding genuine regions at the pixel level. In addition, AR-Net has high robustness on postprocessing operations, such as noise, blur, and JPEG recompression.