In long-term multivariate time–series forecasting, longer input windows are often assumed to help. However, on some standard benchmarks we observe that, for certain model–dataset combinations, performance becomes non-monotonic and unstable as the look-back length grows. Taking a data-mining perspective, we first characterize correlated input noise on the Weather benchmark, a widely used multivariate forecasting dataset. Residual-denoising analyses on this dataset show that, for some variables and time periods, input perturbations exhibit substantial short-lag autocorrelation rather than behaving as white noise. Building on this evidence, we provide a window-conditional second-order interpretation that relates whole-window embeddings to variance inflation (for linear forecasters) and curvature-dependent bias (for nonlinear forecasters) arising from residual lag covariance under temporally correlated perturbations, and discuss how this interpretation connects to both squared and absolute error criteria used in evaluation. To address this issue in real-world datasets, we introduce NoDE (Noise-Decorrelation Embedding), a simple and model-agnostic structured linear embedding that reduces effective lag covariance via downsampling, permutation-based reshaping (shuffle/unshuffle), and shared affine mappings, while preserving a drop-in interface to common backbones. Our protocol automatically selects the downsampling window in a data-driven manner through a two-stage procedure: candidate resolutions are proposed based on ACF-energy statistics, and the final choice is made via a lightweight quick-train validation. An explicit no-op option is included to disable NoDE when downsampling is not beneficial. Comprehensive experimental evaluations validate that NoDE stabilizes performance under long look-back windows across multiple datasets and backbone families (Transformer, CNN, and MLP). In addition, the protocol’s ability to automatically activate or deactivate NoDE via a no-op option allows it to support standard, patch-wise, and multi-scale front-ends through the same selection interface, enabling robust and architecture-agnostic deployment. Code and models are open-source at https://github.com/lmh9507/NoDE .
As an emerging weakly supervised learning paradigm, partial multi-label learning (PML) aims to derive robust predictors from multi-label data with noisy labels. However, apart from label noise, multi-label data usually contain thousands of feature dimensions. Existing PML methods overlook the curse of dimensionality caused by high-dimensional features, which prevents their models from achieving optimal performance. To address the above issues, this paper proposes a novel partial multi-label feature selection method named PMFS-NRDLC, which unifies label noise removal and dimensionality reduction within a single optimization framework. First, by fully utilizing the exclusive prior of PML label noise, we construct a position-constrained regression model with strict zero constraints on noise terms corresponding to non-candidate labels, which fundamentally avoids spurious confidence assignment and achieves precise label denoising. Second, we introduce feature-space manifold topology under the smoothness assumption to regularize label confidence learning, ensuring consistent local structural characteristics between original features and estimated label confidences. Meanwhile, we design an iterative dynamic label graph update strategy to capture time-varying latent label correlations, and further regularize feature weights to guarantee that correlated labels share similar feature importance. Finally, we adopt the l_2,1 -norm to impose sparse constraints on feature weights to implement embedded feature selection. Extensive experiments on various benchmark datasets demonstrate that PMFS-NRDLC consistently outperforms state-of-the-art competitors.
Identifying influential nodes is crucial to understanding and managing the structural and functional dynamics of complex networks. While traditional methods focus on either local or global influence, hybrid approaches aim to integrate both for improved accuracy. This paper presents a novel hybrid method, Triplet Node Influence (TNI), which combines degree centrality, communicability, and gravity-based H-index to evaluate node influence from local, global, and gravitational perspectives. To ensure computational efficiency without sacrificing precision, an optimal truncation radius is introduced. The effectiveness of TNI is evaluated against several state-of-the-art algorithms using three key metrics: (1) SIR-based spreading simulations with Kendall’s tau, (2) the Monotonicity Index, and (3) robustness analysis through node removal. TNI is tested on twelve unweighted and three weighted real-world networks. The results show that TNI consistently demonstrates effectiveness in identifying influential nodes, achieving superior accuracy and strong discriminatory power. It also maintains acceptable network robustness and time performance, offering a more generalized and adaptable node evaluation framework for diverse network types.
As key players of gene expression regulation, microRNAs (miRNAs) are involved in the onset and progression of many diseases. Because of their presence in organic fluid, such as blood or urine, they are optimal candidates as disease biomarkers that can be easily detected in liquid biopsy, which is less invasive and expensive than traditional techniques. In this paper, we propose a novel patient embedding method, called GraphiRNA, to support the diagnosis of Alzheimer’s disease from miRNA expression values. GraphiRNA is capable of capturing information about the relationships among miRNAs, estimated through the correlation of their expression values over a set of patients, and also integrate the information about their RNA sequences. Notably, GraphiRNA is able to properly model negative correlations by introducing synthetic nodes into the network, to avoid discarding the sign of correlations (and the associated information), as done by existing approaches. Our experimental evaluation, performed on real data from three studies on Alzheimer’s disease, showed superior classification performances of GraphiRNA, especially in patients affected by Mild Cognitive Impairment (MCI), which outlines the potential of GraphiRNA of being adopted as a diagnostic tool for the early intervention. Finally, our ablation study on the approach adopted to handle negative correlations showed an improvement reaching 30.9
Multimodal Knowledge Graphs enhance traditional symbolic knowledge graphs by incorporating complementary data such as text and images, enabling richer semantic inference. However, integrating heterogeneous modalities remains challenging due to their distinct geometric and semantic characteristics. We propose CurvBi (Curvature-aware Bicomplex), a novel framework for multimodal knowledge graph completion that unifies modality alignment and relational modeling under a coherent mathematical foundation. CurvBi leverages bicomplex embeddings to capture symmetric, asymmetric, and cyclic relationships in a four-dimensional algebraic space, and applies Riemannian manifold fusion via Fisher-Rao distance to align modality-specific embeddings within a curvature-aware structure. Unlike prior approaches relying on Euclidean projection or shallow concatenation, CurvBi offers a geometry-sensitive and algebraically expressive representation space. We evaluate CurvBi on four benchmark datasets, where it consistently achieves competitive performance across multiple evaluation metrics, with improvements reaching up to 6
