In real-world applications, hierarchical data with complex category structures pose significant challenges due to high dimensionality and potential incompleteness. Existing studies on hierarchical feature selection assume data are complete, without considering incomplete hierarchical data. However, in practice, obtaining a complete and uncorrupted feature space can be challenging due to physical, technical, and cost constraints in data collection. To address this issue, we propose a hierarchical feature selection method for incomplete hierarchical data. First, weight amendment matrices are introduced to shift the primary missing-data compensation from the feature space to the projection-weight space under hierarchical structural guidance. Second, we establish a hierarchical interaction framework that explicitly captures the implicit relationship between the weight amendment matrix and the hierarchical weight matrix via constrained optimization. Third, we design a sparse hierarchical regularization term that adaptively adjusts feature weights based on hierarchy levels. Extensive experiments on benchmark datasets demonstrate the proposed approach’s superior performance over state-of-the-art methods, validating its enhanced effectiveness and progress in addressing the target problem domain.
Snapshot hyperspectral imaging based on a spectral filter array (SFA) enables high-speed acquisition but inherently trades spatial resolution for spectral resolution. Fusing a low-resolution (LR) mosaiced image with a high-resolution (HR) panchromatic (PAN) image offers a promising solution, yet existing methods often exhibit spectral distortion or spatial degradation, especially in unsupervised settings without ground truth. To address this issue, we propose an unsupervised variational posterior learning framework that extends deterministic EI-based fusion into a probabilistic latent-variable formulation. Unlike existing EI-based methods that directly predict a single HR hyperspectral image (HSI) together with a deterministic degradation model, the proposed framework models the HR HSI, spectral response function (SRF), and observation-noise statistics through corresponding posterior distributions. The EI prior is further incorporated as a posterior-consistency constraint between the estimates inferred from original, geometrically transformed, and posterior-predictive observations. Combined with the physical imaging model, this formulation introduces additional stochastic and distribution-level regularization to reduce inference ambiguity without requiring reference HR HSIs. Extensive experiments on both simulated and real-world datasets demonstrate the superiority of the proposed method over the state-of-the-art (SOTA) approaches.
Artificial intelligence (AI)-powered heart sound auscultation offers a noninvasive and accessible approach for diagnosing cardiovascular diseases (CVDs). However, training robust diagnostic models requires large-scale data, which is hindered by strict privacy regulations across isolated medical institutions. While federated learning (FL) addresses privacy concerns, real-world medical data exhibits severe non-independent and non-identically (Non-IID) characteristics, including label heterogeneity, intraclass representation diversity, and interclass imbalance. Existing FL methods primarily focus on personalized client models and struggle to construct a comprehensive global model capable of recognizing all disease subtypes under such extreme data heterogeneity. To overcome these challenges, we propose a novel two-stage decoupled federated prototype learning framework. In the first stage, we introduce a multicluster trainable prototype aggregation mechanism equipped with comprehensive local feature space calibration—incorporating $L_{2}$ normalization, learnable prototype radii, multilevel alignment, and adversarial domain adaptation—to dynamically handle label skew and intraclass diversity. In the second stage, we design a decoupled joint training paradigm that reconstructs a unified global classifier through privacy-preserving one-shot feature sharing and prior-based logit adjustment. Comprehensive experiments on the Yaseen valvular heart diseases (VHDs) and ZCHSound congenital heart diseases (CHDs) datasets demonstrate the advantages of our method. Specifically, our framework effectively narrows extreme performance disparities among individual client models and successfully constructs a robust, reusable global model that outperforms traditional centralized and federated baselines under severe class imbalance, all while ensuring minimal communication overhead.
Given the inherent tradeoff between spectral and spatial resolution, hyperspectral images (HSIs) typically exhibit insufficient spatial details. Combining the HSI with a corresponding high spatial resolution conventional imagery serves as a compromised alternative to produce a high-quality HSI. While fusion-based HSI superresolution (HSR) schemes have demonstrated effectiveness in this task, the spectral variability arising from the fusion of heterogeneous images remains an underexplored challenge in prior research. To this end, this article presents a novel zero-shot fusion model for HSR accounting for spectral variability between fusion materials, where only the test-time inputs are used to learn an image-specific fusion model. The core idea is to represent spectral signatures on a low-dimensional spectral manifold, enabling scene-adaptive modeling of varying endmember (i.e., pure spectral signature) distributions through a two-stream variational network. To further enhance flexibility and robustness, a manifold transformation prior (MTP) is introduced in the latent space to parametrize spectral variations induced by different imaging conditions. The proposed frame jointly estimates varying endmembers and abundance fractions (i.e., subpixel proportions of endmembers) in a unified variational optimization (VO) framework. A comprehensive evaluation on synthetic, semireal, and real-life datasets demonstrates the superior performance and robustness of our method in the HSR task, especially in handling spectral variability.
Multiview data often suffers from missing views, posing significant challenges for incomplete multiview clustering (IMVC). Existing methods typically perform explicit similarity feature-level imputation using observed views, yet they often fail to capture the shared latent distribution across different views, particularly under high missing rates. In addition, many IMVC models adopt a single-stage framework that jointly optimizes data recovery and clustering, resulting in increased model complexity and computational burden due to full-network backpropagation, thus limiting scalability and training efficiency. To address these issues, we propose a two-stage framework, termed hybrid conditional diffusion with implicit repair for IMVC (HCDIR-IMVC), which simplifies the integration of diffusion models while improving the consistency and robustness of the recovered features. In the first stage, we introduce a unified hybrid conditional diffusion restoration (HC-DR) model that leverages probabilistic learning to dynamically select observed views as conditional inputs for the diffusion process. This enables the model to effectively learn the distribution of missing views and guide their generation. To further enhance cross-view consistency, we incorporate contrastive learning into the diffusion framework. Unlike prior approaches that use separate diffusion models for each view, our method adopts a single hybrid-conditioned model, significantly reducing complexity and improving multiview feature utilization. In the clustering stage, we freeze the imputation model’s parameters and remove the decoder to reduce training overhead. A representation and distribution alignment strategy is employed to ensure consistent clustering labels within and across views. Extensive experiments demonstrate that HCDIR-IMVC achieves superior performance in accuracy and computational efficiency, particularly under high missing rates.
