Few-shot class incremental learning (FSCIL) aims to continuously learn new classes from limited training samples while retaining previously acquired knowledge. Existing approaches are not fully capable of balancing stability and plasticity in dynamic scenarios. To overcome this limitation, we introduce a novel FSCIL framework that leverages graph neural networks (GNNs) to model interdependencies between different categories and enhance cross-modal alignment. Our framework incorporates three key components: 1) a Graph Isomorphism Network (GIN) to propagate contextual relationships among prompts; 2) a Hamiltonian Graph Network with Energy Conservation (HGN-EC) to stabilize training dynamics via energy conservation constraints; and 3) an Adversarially Constrained Graph Autoencoder (ACGA) to enforce latent space consistency. By integrating these components with a parameter-efficient CLIP backbone, our method dynamically adapts graph structures to model semantic correlations between textual and visual modalities. Additionally, contrastive learning with energy-based regularization is employed to mitigate catastrophic forgetting and improve generalization. Comprehensive experiments on benchmark datasets validate the framework's incremental accuracy and stability compared to state-of-the-art baselines. This work advances FSCIL by unifying graph-based relational reasoning with physics-inspired optimization, offering a scalable and interpretable framework. Code is available at: https://github.com/aries-yqian/ACHG-CLIP.

Weakly Supervised Object Localization (WSOL) relies only on image-level labels to realize object localization, significantly reducing the cost for fine-grained annotations. While traditional CAM-based methods excel at identifying the most prominent regions of objects, they frequently neglect other essential components, resulting in partial or incomplete object localization. The foreground prediction map (FPM) generates finer-grained activation maps using underlying features to address the shortcomings of CAM, but it may still have coverage blind spots. To this end, this paper proposes a collaborative optimization framework based on cross-modal semantic alignment that deeply integrates the saliency awareness of CAM with the refined representation capabilities of FPM. It introduces a multimodal pretrained model (CLIP) to construct a semantic-driven WSOL paradigm. By dynamically interacting CLIP's text embeddings with the semantic of image categories, a semantic-enhanced FPM based on similarity measurement is generated. Leveraging CLIP's cross-modal alignment capabilities, a targeted generation scheme is designed. On the one hand, the CLIP model is frozen and its features are refined through a decoder to obtain richer semantic representations; On the other hand, by using knowledge distillation, the CAM generated by CLIP is taken as a reference benchmark, guiding the network to learn more accurate target localization. Additionally, to enhance FPM's focus on foreground regions, the Exponential Decay Foreground Emphasis (EDFE) module is designed, which uses a differentiated excitation strategy to effectively suppress background interference and highlight target areas. Experimental results show that our method significantly improves the completeness and boundary accuracy of target localization under weak supervision, laying a solid foundation for subsequent downstream tasks.

ABSTRACT To achieve spatiotemporal continuity or some sparsity for robust tracking, most current discriminative correlation filter (DCF) methods introduce new regularization terms or self‐adaption hyperparameters to restrict the trackers. However, regardless of the validity of the pseudo‐Gaussian label, previous DCF trackers generally suffer from aberrance, mismatching. In this work, we rethink the DCF tracker from the label matching and propose a label approximation DCF tracker (LACF) focusing on analyzing the commonly used Gaussian pseudo labels in the DCF. Specifically, based on the assumption that the same objects should contain a similar response between two frames, we construct a new pseudo label that combines the original pseudo‐Gaussian labels and the previous response map. On the other hand, we introduce a windowing strategy to focus the DCF model on matching crucial labels for the right position. The experimental results demonstrate that LACF significantly achieves competitive performance for real‐time CPU small unmanned aircraft tracking.

Fuzzy mathematical theory is widely used, fuzzy optimization is a branch of fuzzy mathematical theory, the significant application area is artificial intelligence in computer science, especially machine learning (deep learning) and pattern recognition. Fuzzy mathematics, especially fuzzy optimization, has become a bridge between the manifold optimization theory and deep learning applications, which is an essential theoretical foundation. The manifold optimization algorithm employs the projection method, which is unstable. In order to resolve the problem, in this article, the theory and methodology of manifold optimization concerning real and complex spaces is fully considered. Our primary focus is on the Riemannian manifold, where a groundbreaking optimization algorithm with the dual function and a fuzzy valuation step is proposed. To accelerate the convergence and enhance the stability of the optimization algorithm, a novel learning rate is present, which is referred as bivariate gradual learning rate warm-up. A comprehensive analysis of its convergence rates is conducted in various scenarios and the experiments results substantiate our discoveries, and demonstrate the correctness and effectiveness of our devised algorithm.

Graph neural networks (GNNs) could directly deal with the data of graph structure. Current GNNs are confined to the spatial domain and learn real low-dimensional embeddings in graph classification tasks. In this article, we explore frequency domain-oriented complex GNNs in which the node's embedding in each layer is a complex vector. The difficulty lies in the design of graph pooling and we propose a mirror-connected design with two crucial problems: parameter reduction problem and complex gradient backpropagation problem. To deal with the former problem, we propose the notion of squared singular value pooling (SSVP) and prove that the representation power of SSVP followed by a fully connected layer with nonnegative weights is exactly equivalent to that of a mirror-connected layer. To resolve the latter problem, we provide an alternative feasible method to solve singular values of complex embeddings with a theoretical guarantee. Finally, we propose a mixture of pooling strategies in which first-order statistics information is employed to enrich the last low-dimensional representation. Experiments on benchmarks demonstrate the effectiveness of the complex GNNs with mirror-connected layers.

Contrastive learning (CL) is a popular learning paradigm in deep learning, which uses contrastive principle to learn low-dimensional embeddings, and has been applied in Graph Neural Networks (GNNs) successfully. Existing works of contrastive multi-view GNNs usually focus on point-to-point contrastive learning strategies. However, they neglect the local information in neighbors, which brings isolated positive samples. The quality of selected positive samples is hard to evaluate, and these samples may lead to invalid contrastiveness. Therefore, we propose a simple and efficient neighbors-to-neighbors contrastive graph neural network (NNC-GCN), which constructs a consistent multi-view by using the topologies of original input graphs. Moreover, we raise a new learning problem of unlabeled data base on these constructed multi-view topologies and propose a loss function NNC-InfoNCE to guide its learning process. The NNC-InfoNCE is an improved version of InfoNCE, which can be adapted to neighborhood-level contrast learning. Specifically, the neighborhoods and the remaining nodes of the selected anchor are weighted and treated as positive and negative sample sets. The experimental results show that our method is effective on public benchmark datasets.

Multiview clustering (MVC), which can dexterously uncover the underlying intrinsic clustering structures of the data, has been particularly attractive in recent years. However, previous methods are designed for either complete or incomplete multiview only, without a unified framework that handles both tasks simultaneously. To address this issue, we propose a unified framework to efficiently tackle both tasks in approximately linear complexity, which integrates tensor learning to explore the inter-view low-rankness and dynamic anchor learning to explore the intra-view low-rankness for scalable clustering (TDASC). Specifically, TDASC efficiently learns smaller view-specific graphs by anchor learning, which not only explores the diversity embedded in multiview data, but also yields approximately linear complexity. Meanwhile, unlike most current approaches that only focus on pair-wise relationships, the proposed TDASC incorporates multiple graphs into an inter-view low-rank tensor, which elegantly models the high-order correlations across views and further guides the anchor learning. Extensive experiments on both complete and incomplete multiview datasets clearly demonstrate the effectiveness and efficiency of TDASC compared with several state-of-the-art techniques.

Ensemble clustering (EC), utilizing multiple basic partitions (BPs) to yield a robust consensus clustering, has shown promising clustering performance. Nevertheless, most current algorithms suffer from two challenging hurdles: (1) a surge of EC-based methods only focus on pair-wise sample correlation while fully ignoring the high-order correlations of diverse views. (2) they deal directly with the co-association (CA) matrices generated from BPs, which are inevitably corrupted by noise and thus degrade the clustering performance. To address these issues, we propose a novel Double High-Order Correlation Preserved Robust Multi-View Ensemble Clustering (DC-RMEC) method, which preserves the high-order inter-view correlation and the high-order correlation of original data simultaneously. Specifically, DC-RMEC constructs a hypergraph from BPs to fuse high-level complementary information from different algorithms and incorporates multiple CA-based representations into a low-rank tensor to discover the high-order relevance underlying CA matrices, such that double high-order correlation of multi-view features could be dexterously uncovered. Moreover, a marginalized denoiser is invoked to gain robust view-specific CA matrices. Furthermore, we develop a unified framework to jointly optimize the representation tensor and the result matrix. An effective iterative optimization algorithm is designed to optimize our DC-RMEC model by resorting to the alternating direction method of multipliers. Extensive experiments on seven real-world multi-view datasets have demonstrated the superiority of DC-RMEC compared with several state-of-the-art multi-view ensemble clustering methods.

Although the Siamese tracker has evolved rapidly in recent years, lacking discrimination has always been its biggest disadvantage, compared to the discriminative correlation filters (DCF) tracker. It is because mainstream Siamese trackers take only the initial frame as the template. In this paper, we present a new multi-template transformer-based tracker to harness the tracking target’s temporal features and distinguish them from several distractors. It introduces the cache algorithm to adaptively store templates and dynamically recommend the most appropriate templates for the tracker. Moreover, a multi-template integration method is proposed here, by designing a transformer block with learnable template embedding, to reduce the impact of drifts in dynamic templates. We conduct quantitative experiments on multi-distractor scenarios and general scenarios. The results show the proposed tracker outperforms state-of-the-art trackers in multi-distractor scenarios, and can still keep real-time speed. Codes and models are available on GitHub1.

Zero-shot learning (ZSL) aims to identify unseen classes with zero samples during training. Broadly speaking, present ZSL methods usually adopt class-level semantic labels and compare them with instance-level semantic predictions to infer unseen classes. However, we find that such existing models mostly produce imbalanced semantic predictions, i.e. these models could perform precisely for some semantics, but may not for others. To address the drawback, we aim to introduce an imbalanced learning framework into ZSL. However, we find that imbalanced ZSL has two unique challenges: (1) Its imbalanced predictions are highly correlated with the value of semantic labels rather than the number of samples as typically considered in the traditional imbalanced learning; (2) Different semantics follow quite different error distributions between classes. To mitigate these issues, we first formalize ZSL as an imbalanced regression problem which offers empirical evidences to interpret how semantic labels lead to imbalanced semantic predictions. We then propose a re-weighted loss termed Re-balanced Mean-Squared Error (ReMSE), which tracks the mean and variance of error distributions, thus ensuring rebalanced learning across classes. As a major contribution, we conduct a series of analyses showing that ReMSE is theoretically well established. Extensive experiments demonstrate that the proposed method effectively alleviates the imbalance in semantic prediction and outperforms many state-of-the-art ZSL methods.

This paper addresses the multi-view subspace clustering problem and proposes the self-paced enhanced low-rank tensor kernelized multi-view subspace clustering (SETKMC) method, which is based on two motivations: (1) singular values of the representations and multiple instances should be treated differently. The reasons are that larger singular values of the representations usually quantify the major information and should be less penalized; samples with different degrees of noise may have various reliability for clustering. (2) many existing methods may cause the degraded performance when multi-view features reside in different nonlinear subspaces. This is because they usually assumed that multiple features lie within the union of several linear subspaces. SETKMC integrates the nonconvex tensor norm, self-paced learning, and kernel trick into a unified model for multi-view subspace clustering. The nonconvex tensor norm imposes different weights on different singular values. The self-paced learning gradually involves instances from more reliable to less reliable ones while the kernel trick aims to handle the multi-view data in nonlinear subspaces. One iterative algorithm is proposed based on the alternating direction method of multipliers. Extensive results on seven real-world datasets show the effectiveness of the proposed SETKMC compared to fifteen state-of-the-art multi-view clustering methods.

Covariate shift assumption based domain adaptation approaches usually utilize only one common transformation to align marginal distributions and make conditional distributions preserved. However, one common transformation may cause loss of useful information, such as variances and neighborhood relationship in both source and target domain. To address this problem, we propose a novel method called homologous component analysis (HCA) where we try to find two totally different but homologous transformations to align distributions with side information and make conditional distributions preserved. As it is hard to find a closed form solution to the corresponding optimization problem, we solve them by means of the alternating direction minimizing method (ADMM) in the context of Stiefel manifolds. We also provide a generalization error bound for domain adaptation in semi-supervised case and two transformations can help to decrease this upper bound more than only one common transformation does. Extensive experiments on synthetic and real data show the effectiveness of the proposed method by comparing its classification accuracy with the state-of-the-art methods and numerical evidence on chordal distance and Frobenius distance shows that resulting optimal transformations are different.

Domain adaptation has proven to be successful in dealing with the case where training and test samples are drawn from two kinds of distributions, respectively. Recently, the second-order statistics alignment has gained significant attention in the field of domain adaptation due to its superior simplicity and effectiveness. However, researchers have encountered major difficulties with optimization, as it is difficult to find an explicit expression for the gradient. Moreover, the used transformation employed here does not perform dimensionality reduction. Accordingly, in this article, we prove that there exits some scaled LogDet metric that is more effective for the second-order statistics alignment than the Frobenius norm, and hence, we consider it for second-order statistics alignment. First, we introduce the two homologous transformations, which can help to reduce dimensionality and excavate transferable knowledge from the relevant domain. Second, we provide an explicit gradient expression, which is an important ingredient for optimization. We further extend the LogDet model from single-source domain setting to multisource domain setting by applying the weighted Karcher mean to the LogDet metric. Experiments on both synthetic and realistic domain adaptation tasks demonstrate that the proposed approaches are effective when compared with state-of-the-art ones.

Subspace clustering aims to segment data sampled from a union of subspaces in visual data tasks. Structured Sparse Subspace Clustering (SSSC) model is a unified optimization framework, which proves successful in learning both the self representation of the data and their subspace segmentation. However, SSSC involves solving non-convex subproblems and hence it may be stuck into bad local minima such that clustering performance degrades. In this paper, we propose a self-paced subspace clustering algorithm to tackle this problem, which learns subspace segmentation of data by progressing from 'easy' to 'complex' examples under a novel self-paced regularizer. Experiments on the real-world human face datasets verify the effectiveness of the proposed algorithm.