Local community detection in multiplex networks faces two critical challenges. First, the result is sensitive to the query seed position. A core seed yields more accurate detection than a peripheral one. Second, nodes near community boundaries often have ambiguous memberships, making their community assignment difficult. These issues limit the accuracy and robustness of existing methods. To address these issues, this paper proposes MLCD-CND (Local Community Detection in Multiplex Networks via Core Node Detection), a two-stage method for local community detection in multiplex networks. In the first stage, MLCD-CND mitigates seed dependency by detecting high-quality core nodes through a novel Node Relation Strength metric, which integrates cross-layer structural similarity (via Jaccard Similarity Coefficient) and interaction density (via Node Influence Area Density). These core nodes replace the initial query seed to form a robust initialization. In the second stage, the algorithm expands the community by optimizing a hybrid metric that balances multiplex local modularity and community relation strength, effectively preventing over-expansion into sparse regions. The experimental results demonstrate that MLCD-CND improves the performance of the F-score by up to 72% compared to the baseline model. The method effectively addresses seed dependency and over-expansion, offering robust performance across diverse networks.
Boolean matrix factorization is a popular technique for summarizing a binary matrix. The goal in such summary is to find k tiles that are denser than the rest of the data: a typical objective is to minimize the number of (weighted) disagreements with the input data. Another instance of minimizing weighted disagreements is correlation clustering, where we are given a graph with signed edges and are asked to partition a graph minimizing the negative intra edges and the positive inter edges. We argue that weighting disagreements in both cases is crucial, especially for sparse datasets. This leads to a problem of selecting an appropriate weight: a non-trivial problem as we cannot use the original objective directly. To solve this issue we propose an alternative strategy: we search for results that maximize the likelihood of a simple model, based on two Bernoulli variables. We then show that we can find good solutions using a grid search. We develop 2 grid search strategies. The first approach requires 𝒪( (1 + ϵ ^-1) logℓ) queries, where ℓ is the size of the input, and yields (1 + ϵ )α guarantee if we can α -approximate the original weighted disagreement problem. The second approach is a pruning technique, based on fractional programming, that reduces the number of queries significantly in practice, especially for small ϵ . This approach yields (1 + ϵ ) approximation guarantees, if we can solve the original weighted disagreement problem exactly. We demonstrate empirically that weighting disagreements produces more accurate results, and study the cases where the grid search yields better likelihood scores than the selected baselines.
Multi-label feature selection (MFS) is vital for high-dimensional biological data analysis, where samples such as genes or proteins are annotated with multiple functional labels. A major challenge in this setting is that MFS methods must identify informative features while simultaneously modeling complex, and often noisy,inter-label dependencies. Existing MFS methods, however, suffer from three critical limitations in real-world biological applications: (i) prevalent annotation noise, which compromises model reliability; (ii) limited integration of domain-specific knowledge, leading to biologically implausible or spurious label dependencies; and (iii) instability in hierarchical dependency modeling, which propagates errors across prediction stages. To address these challenges, we propose Self-Distilled Label Graph Learning for Adaptive Multi-Label Feature Selection (SDMFS), a unified optimization framework that simultaneously performs feature selection and adaptive label graph learning. First, a self-distillation mechanism generates temporally smoothed teacher targets from the model’s own historical predictions, mitigating annotation noise and stabilizing optimization. Second, an adaptive prior-guided label graph learning module integrates empirical label prior knowledge with data-driven label dependency discovery, ensuring the learned structures are both reliable and interpretable. Finally, the framework jointly leverages these modules to select discriminative features that preserve functional relationships across labels. We develop an efficient alternating minimization algorithm with guaranteed convergence and demonstrate, through extensive experiments on seven benchmark biological datasets, that SDMFS consistently outperforms state-of-the-art MFS methods in predictive performance, robustness, and biological interpretability.
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.
