Generalized Category Discovery (GCD) aims to classify unlabeled data by leveraging knowledge from labeled categories. While existing methods have achieved remarkable progress, they often treat images as flat feature sets, neglecting the intrinsic hierarchy: where key objects dominate meaning and backgrounds serve as context. For instance, in images of a dog either standing on grass or lying on a bed, the dog remains the central semantic element, whereas the background varies. Motivated by this, we propose LEArning Intrinsic Hierarchy (LEAH), a lightweight plug-and-play module designed to model hierarchical structure within images. LEAH consists of two components: a pruner that filters task-irrelevant tokens to extract key objects, and a constructor that embeds key objects and full images into hyperbolic space using adaptive entailment cones to capture compositional semantics. LEAH can be easily integrated into existing GCD frameworks with minimal modification. When applied to SimGCD, it achieves up to 13.2% accuracy improvement on fine-grained benchmarks, demonstrating its effectiveness in discovering subtle inter-class differences through hierarchical modeling.

Artificial intelligence-generated content (AIGC) aims to represent new content synthesized by learning rules from the existing data and has expanded its applications to the health domain, particularly generating medical robotic surgical scenes. Most pioneering methods rely primarily on 2D representations and thus will inevitably suffer from scene distortion when large surgical tool movements and intense soft tissue deformation are encountered. Recent works instead employ explicit 3D structural representations or implicit neural rendering to improve performance under large pose changes. Nevertheless, the fidelity of structure and texture is not so desirable, especially for novel-view synthesis. In this paper, we propose H-RSSG(high-fidelity robotic surgical scene generation with an implicit deformable neural radiance field), which achieves high-fidelity and free-view robotic-surgical-scenes synthesis. The H-RSSG explores the potential of artificial intelligence in generating high-fidelity medical robotic surgical scenes. In addition to generating robotic surgical scenes, the H-RSSG provides standardized evaluation metrics by establishing a benchmark dataset. Moreover, H-RSSG enhances the robustness of AIGC-generated scenes by combining adjacent view feature fusion and multi-view correspondence loss to improve spatial context consistency. A notable advantage of the H-RSSG is its integration of an extended transform-based depth perception module with a lightweight neural network used to estimate masking depths even in the presence of surgical instruments. This results in spatiotemporally consistent high-fidelity robotic surgical scenes. The results of the experiments conducted on the self-established and EndoNeRF datasets have a PRMSE of about 0.0341, an average improvement of 12.1% in PSNR, and an average improvement of 6.69% in SSIM, which demonstrates that the proposed H-RSSG can outperform the benchmark models’ performance, achieving state-of-the-art results. This study highlights the AIGC’s potential to create highly realistic surgical training environments for medical professionals and applications in the medical industry.Video is available at:H-RSSG.

Generalized Category Discovery (GCD) aims to classify unlabeled data from both known and unknown categories by leveraging knowledge from labeled known categories. While existing methods have made notable progress, they often overlook a hidden stumbling block in GCD: distracted attention. Specifically, when processing unlabeled data, models tend to focus not only on key objects in the image but also on task-irrelevant background regions, leading to suboptimal feature extraction. To remove this stumbling block, we propose Attention Focusing (AF), an adaptive mechanism designed to sharpen the model's focus by pruning non-informative tokens. AF consists of two simple yet effective components: Token Importance Measurement (TIME) and Token Adaptive Pruning (TAP), working in a cascade. TIME quantifies token importance across multiple scales, while TAP prunes non-informative tokens by utilizing the multi-scale importance scores provided by TIME. AF is a lightweight, plug-and-play module that integrates seamlessly into existing GCD methods with minimal computational overhead. When incorporated into one prominent GCD method, SimGCD, AF achieves up to 15.4% performance improvement over the baseline with minimal computational overhead. The implementation code is provided in https://github.com/Afleve/AFGCD.

Most existing multi-view graph clustering models focus on integrating the topological structure of different views directly, which cannot efficiently stimulate the collaboration between multiple views. To alleviate this problem, this paper proposes a Triangle Topology Enhancement (T2E) module, which expands two topological structures based on the raw topology of each view, including the self-triangle enhanced topology that highlights the local view information and the cross-view triangle enhanced topology containing the global-local view information. Afterward, this paper designs a novel multi-view graph clustering model, named MGC-T2E, to integrate both the raw and derived topological structures and directly induce consistent clustering indicators based on a self-supervised clustering module. In the simulation, the experimental results demonstrate that MGC-T2E achieves state-of-the-art performances compared with a mass of current competitors.

This study addresses the problem of generalized category discovery (GCD), an advanced and challenging semi-supervised learning scenario that deals with unlabeled data from both known and novel categories. Although recent research has effectively engaged with this issue, these studies typically map features into Euclidean space, which fails to maintain the latent semantic hierarchy of the training samples effectively. This limitation restricts the exploration of more detailed and rich information and degrades the performance in discovering new categories. The emerging field of hyperbolic representation learning suggests that hyperbolic geometry could be advantageous for extracting semantic information to tackle this problem. Motivated by this, we proposed hyperbolic hierarchical representation learning for GCD (HypGCD). Specifically, HypGCD enhances representations in hyperbolic space, building upon the Euclidean space representation from two perspectives: instance-class level and instance-instance level. At the instance-class level, HypGCD endeavors to construct well-defined clusters, with each sample forming a robust hierarchical cluster structure. Concurrently, at the instance-instance level, HypGCD anticipates that a subset of samples will display a tree-like structure in local space, which aligns more closely with real-world scenarios. Finally, HypGCD optimizes the Euclidean and hyperbolic space collectively to obtain refined features. Additionally, we show that HypGCD is exceptionally effective, achieving state-of-the-art (SOTA) results on several datasets. The code is available at https://github.com/DuannYu/HypGCD.

Clustering-based methods are emerging as a ubiquitous technology in unsupervised object Re-Identification (ReID), which alternate between pseudo-label generation and representation learning. Recent advances in this field mainly fall into two groups: pseudo-label correction and robust representation learning. Differently, in this work, we improve unsupervised object ReID from feature calibration, a completely different but complementary insight from the current approaches. Specifically, we propose to insert a conceptually simple yet empirically powerful Feature Calibration Module (FCM) before pseudo-label generation. In practice, FCM calibrates the features using a nonparametric graph attention network, enforcing similar instances to move together in the feature space while allowing dissimilar instances to separate. As a result, we can generate more reliable pseudo-labels using the calibrated features and further improve subsequent representation learning. FCM is simple, effective, parameter-free, training-free, plug-and-play, and can be considered as a catalyst, increasing the ’chemical reaction’ between pseudo-label generation and representation learning. Moreover, it maintains the efficiency of testing time with negligible impact on training time. In this paper, we insert FCM into a simple baseline. Experiments across different scenarios and benchmarks show that FCM consistently improves the baseline (e.g., 8.2% mAP gain on MSMT17), and achieves the new state-of-the-art results. Code is available at: https://github.com/lhf12278/FCM-ReID.

Continuous affect recognition from facial images aims to estimate the values of multiple affective dimensions from a facial image sequence. To leverage relevant information between multiple affective dimensions, multitask learning has been used in the estimation of continuous affective states . Most of the existing multitask continuous affect recognition methods focus on designing elaborate multitask networks. Meanwhile, a few research works consider using multitask training strategies for continuous affect recognition. In general, existing multitask continuous affect recognition methods face the problem of unstable training effects. In this work, to improve the stability of multitask learning , we propose an ensemble learning-enhanced multitask network architecture for continuous affect recognition. In addition, we introduce a novel adaptive weighted loss-based multitask learning strategy to effectively train the proposed multitask continuous affect recognition model. Experimental results, on the RECOLA, SEMAINE and AFEW-VA datasets for continuous affect recognition, demonstrate the potential of the proposed method compared to state-of-the-art methods.

New Intent Discovery (NID) is the task of categorizing new and unknown intents into distinct clusters. Recent advances in this issue can be roughly grouped into parametric clustering and representation learning. Although both of them have demonstrated exceptional performance, the efficacy of combining these two techniques remains unexplored. To this end, we propose a new framework named INS-NID (INteractiveSupervision forNewIntentDiscovery). This framework is designed to build a bridge between parametric clustering and representation learning. In practice, INS-NID comprises a parametric clustering branch and a representation learning branch, which work collaboratively and provide interactive supervision to boost each other. Specifically, the representation learning branch provides reliable distribution estimation during training, which is used to regularize the cluster assignments of the parametric clustering. On the other hand, the cluster assignments predicted by parametric clustering provide additional supervision information for the representation learning branch. The effectiveness of INS is highlighted with superior performance over several state-of-the-art methods across various benchmarks. For example, INS achieves 90% clustering accuracy on the Banking dataset, surpassing the best competitor by 5.96%.

Deep clustering seeks to group data into distinct clusters using deep learning techniques . Existing approaches of deep clustering can be broadly categorized into two groups: offline clustering based on unsupervised representation learning and online clustering based on unsupervised classification . While both groups have demonstrated impressive performance in deep clustering, no study has explored the integration of their respective strengths. To this end, we propose Mutual-Taught Deep Clustering (MTDC), which unifies unsupervised representation learning and unsupervised classification into a framework while realizing mutual promotion using a novel mutual-taught mechanism . Specifically, MTDC alternates between predicting pseudolabels in label space and estimating semantic similarity in feature space during training. Moreover, pseudolabels provide weakly-supervised information to enhance unsupervised representation learning, while semantic similarities function as structural priors that regularize unsupervised classification. Consequently, unsupervised classification and unsupervised representation learning can mutually benefit from one another. MTDC is decoupled from prevailing deep clustering methods . For the sake of clarity, we build upon a straightforward baseline in this paper. Despite its simplicity, we demonstrate that MTDC is exceedingly efficacious and consistently enhances the baseline results by substantial margins. For example, MTDC achieves 2.5 % ∼ 7.9 % (NMI), 3.0 % ∼ 13.9 % (ACC), and 3.1 % ∼ 16.7 % (ARI) gains over the baseline on six widely used image datasets. Source code is available at:https://github.com/yichenwang231/MTDC.

This work focuses on the task of Video-based Visible-Infrared Person Re-Identification, a promising technique for achieving 24-hour surveillance systems. Two main issues in this field are modality discrepancy mitigating and spatial–temporal information mining. In this work, we propose a novel method, named Intermediary-guided Bidirectional spatial–temporal Aggregation Network (IBAN), to address both issues at once. Specifically, IBAN is designed to learn modality-irrelevant features by leveraging the anaglyph data of pedestrian images to serve as the intermediary. Furthermore, a bidirectional spatial–temporal aggregation module is introduced to exploit the spatial–temporal information of video data, while mitigating the impact of noisy image frames. Finally, we design an Easy-sample-based loss to guide the final embedding space and further improve the model’s generalization performance. Extensive experiments on Video-based Visible-Infrared benchmarks show that IBAN achieves promising results and outperforms the state-of-the-art ReID methods by a large margin, improving the rank-1/mAP by$1.29\%/3.46\%$at the Infrared to Visible situation, and by$5.04\%/3.27\%$at the Visible to Infrared situation. The source code of the proposed method will be released athttps://github.com/lhf12278/IBAN.

In the field of data mining, how to deal with high-dimensional data is a fundamental problem. If they are used directly, it is not only computationally expensive but also difficult to obtain satisfactory results. Unsupervised feature selection is designed to reduce the dimension of data by finding a subset of features in the absence of labels. Many unsupervised methods perform feature selection by exploring spectral analysis and manifold learning, such that the intrinsic structure of data can be preserved. However, most of these methods ignore a fact: due to the existence of noise features, the intrinsic structure directly built from original data may be unreliable. To solve this problem, a new unsupervised feature selection model is proposed. The graph structure, feature weights, and projection matrix are learned simultaneously, such that the intrinsic structure is constructed by the data that have been feature weighted and projected. For each data point, its nearest neighbors are acquired in the process of graph construction. Therefore, we call them adaptive neighbors. Besides, an additional constraint is added to the proposed model. It requires that a graph, corresponding to a similarity matrix, should contain exactly c connected components. Then, we present an optimization algorithm to solve the proposed model. Next, we discuss the method of determining the regularization parameter γ in our proposed method and analyze the computational complexity of the optimization algorithm. Finally, experiments are implemented on both synthetic and real-world datasets to demonstrate the effectiveness of the proposed method.

Among a big body of recently developed algorithms for machine learning and data mining, a class of models using non-convex/non-smooth sparsity-inducing norms achieves promising results on many challenging tasks. An important problem faced with such models is to find an effective solution for the objective function with one or multiple intractable terms. Although a large number of optimization approaches have been developed, most of them are tailored to a specific model. Besides, these approaches generally introduce some additional parameters and no longer guarantee convergence. In this work, we first revisit some representative non-convex/non-smooth machine learning models, and then unity them into a generic formulation. Theoretically, we develop a simple yet efficient optimization framework, namely Iteratively Re-Weighted method (IRW), to solve such a class of models and provide the corresponding convergence analysis. Particularly, we validate our proposed method on two challenging machine learning tasks: multi-task regression and feature selection. Source codes are available at:https://github.com/KDD-Code/Sparse.git.

This work focuses on the projected clustering problem. Specifically, an efficient and parameter-free clustering model, named discriminative projected clustering (DPC), is proposed for simultaneously low-dimensional and discriminative projection learning and clustering, from the perspective of least squares regression. The proposed DPC, a constrained regression model, aims at finding both a transformation matrix and a binary indicator matrix to minimize the sum-of-squares error. Theoretically, a significant conclusion is drawn and used to reveal the connection between DPC and linear discriminant analysis (LDA). Experimentally, experiments are conducted on both toy and real-world data to validate the effectiveness and efficiency of DPC; experiments are also conducted on hyperspectral images to further verify its practicability in real-world applications. Experimental results demonstrate that DPC achieves comparable or superior results to some state-of-the-art clustering methods.

Matrix completion, in essence, involves recovering a low-rank matrix from a subset of its entries. Most existing methods for matrix completion neglect two significant issues. First, in several practical applications, such as collaborative filtering, some samples may be corrupted completely. However, most of the robust algorithms consider only the condition that a few components of each column have been arbitrarily contaminated. Second, many real data are not static in nature. Nevertheless, the conventional batch-based matrix completion methods cannot efficiently deal with the out-of-sample, that is, the vector completion problem. In this article, we first provide a novel robust matrix completion model and then develop an efficient optimization method that only requires conducting one time singular value decomposition for a thin matrix per iteration. Furthermore, by exploiting the essence of online matrix completion algorithms, we develop a vector completion model which can help users predict the missing values of out of sample. Numerical comparisons with traditional batch-based and online matrix completion algorithms demonstrate the benefits of the proposed method on streaming data corrupted by column outliers. Moreover, we show that our model can be used to detect outliers from incomplete information. This advantage is validated via numerous experimental results on synthetic and real-world data.

Seismic interpretation is often limited by low resolution and strong noise data. To deal with this issue, we propose to leverage deep convolutional neural network (CNN) to achieve seismic image super-resolution and denoising simultaneously. To train the CNN, we simulate a lot of synthetic seismic images with different resolutions and noise levels to serve as training data sets. To improve the perception quality, we use a loss function that combines the$\ell _{1}$loss and multiscale structural similarity loss. Extensive experimental results on both synthetic and field seismic images demonstrate that the proposed workflow can significantly improve the perception of quality of original data. Compared to conventional methods, the network obtains better performance in enhancing detailed structural and stratigraphic features, such as thin layers and small-scale faults. From the seismic images super-sampled by our CNN method, a fault detection method can compute more accurate fault maps than from the original seismic images.

In this paper, a novel low-rank structural model is proposed for segmenting data drawn from a high-dimensional space. Our method is based on the fact that all groups clustered from a high-dimensional dataset are distributed in multiple low-rank subspaces. In general, it’s a very difficult task to find the low-rank structures hidden in data. Different from the classical sparse subspace clustering (SSC) and low-rank representation (LRR) which all take two steps including building the affinity matrix and spectral clustering, we introduce a new rank constraint into our model. This constraint allows our model to learn a subspace indicator which can capture different clusters directly from the data without any postprocessing. To further approximate the rank constraint, a piecewise function is utilized as the relaxing item for the proposed model. Besides, under the subspace indicator constraints, the integer programming problem is avoided, which makes our algorithm more efficient and scalable. In addition, we prove the convergence of the proposed algorithm in theory and further discuss the general case in which subspaces don’t pass through the origin. Experiment results on both synthetic and real-world datasets demonstrate that our algorithm significantly outperforms the state-of-the-art methods.

Without any prior structure information, Nuclear Norm Minimization (NNM), a convex relaxation for Rank Minimization (RM), is a widespread tool for matrix completion and relevant low rank approximation problems. Nevertheless, the result derivated by NNM generally deviates the solution we desired, because NNM ignores the difference between different singular values. In this paper, we present a non-convex regularizer and utilize it to construct two matrix completion models. In order to solve the constructed models efficiently, we develop an efficient optimization method with convergence guarantee, which can achieve faster convergence speed compared to conventional approaches. Particularly, we show that the proposed regularizer as well as optimization method are suitable for other RM problems, such as subspace clustering based on low rank representation. Extensive experimental results on real images demonstrate that the constructed models provide significant advantages over several state-of-the-art matrix completion algorithms. In addition, we implement numerous experiments to investigate the convergence speed of developed optimization method.

Robust low rank approximation is central to many computer version and data mining domains. Although numerous algorithms have been developed to cope with this issue, most of them considered only the setting that the input data matrix is contaminated by sparse noise and ignored the existing of column-outliers. Here, the outliers represent the columns corrupted by noise completely. To recover the low rank component of a data matrix contaminated by sparse noise and outliers simultaneously, in this paper we first implement a novel robust low rank approximation model to recover a low rank matrix L, and then utilize the subspace obtained by conducting SVD on L to estimate the low rank component of residual data. Numer-ical experiments on real data and synthetic data demonstrate the superiority of proposed method.

This paper proposes a novel algorithm, named Non-Convex Calibrated Multi-Task Learning (NC-CMTL), for learning multiple related regression tasks jointly. Instead of utilizing the nuclear norm, NC-CMTL adopts a non-convex low rank regularizer to explore the shared information among different tasks. In addition, considering that the regularization parameter for each regression task desponds on its noise level, we replace the least squares loss function by square-root loss function. Computationally, as proposed model has a nonsmooth loss function and a non-convex regularization term, we construct an efcient re-weighted method to optimize it. Theoretically, we frst present the convergence analysis of constructed method, and then prove that the derived solution is a stationary point of original problem. Particularly, the regularizer and optimization method used in this paper are also suitable for other rank minimization problems. Numerical experiments on both synthetic and real data illustrate the advantages of NC-CMTL over several state-of-the-art methods.