Although the research of traditional multi-view clustering has made great progress, most methods still require ideally constructed datasets as input. However, in some real applications, the multi-view data may be partially aligned, leading to the problem of partially view-aligned clustering. Existing partially view-aligned clustering methods rely on numerous and expensive alignment information to obtain the promising results. Thus, we propose a novel method, termed partially view-aligned clustering with anchor graph (AGPVC) to tackle this problem, which can adapt to less alignment scenario favorably. Specifically, AGPVC employs anchors to reconstruct the inter-view alignment relationship, and further explores the within-view and cross-view consistency of correspondence to reduce the unreliable alignment in category-level. In this way, reliable alignment across views can be obtained. Different from the existing post-concatenation methods, the reconstructed correspondences are exploited to constrain the recovered data by decoders, thus AGPVC learns a fusion representation for each view, which can integrate the information of other views while satisfy the alignment relationship. Experimental results on several real-world multi-view datasets confirm its superiority compared to other state-of-the-art methods for partially view-aligned clustering.

Multivariate time series forecasting is a fundamental research problem in the Internet of Things (IoT), as it provides critical decision support and underpins the accuracy and reliability of downstream intelligent systems. Existing approaches commonly rely on sequence decomposition strategies that separate time series into trend, seasonal, and residual components. However, limited attention has been paid to how temporal information can be represented and integrated across different granularities. Inspired by the human reading process, in which information is progressively understood at the word, sentence, and paragraph levels, we propose LXformer, a novel forecasting framework that captures multiscale temporal representations. After segmenting multivariate time series into patches, LXformer integrates information at multiple granularities by modeling intra-patch features, inter-patch dependencies, and inter-sequence relationships to accomplish the forecasting task. Specifically, multiple one-dimensional convolutional branches are employed to extract fine-grained local patterns within each patch from diverse perspectives. In addition, agent attention is introduced to facilitate effective interactions across patches and channels, enabling the modeling of coarser-grained temporal dependencies. The combination of one-dimensional convolutions and linear-complexity attention mechanisms ensures that LXformer maintains overall linear computational complexity. Extensive experiments conducted on nine large-scale real-world datasets demonstrate that LXformer consistently achieves lower forecasting errors while delivering faster inference speed and reduced memory consumption. These advantages make LXformer particularly suitable for deployment on edge devices with limited computational resources but high accuracy requirements.

Real-life data is generally represented by multi-view format because multi-view data can describe the charac-teristics of samples well through complementary information. However, when fusing complementary information, we cannot guarantee the alignment between the samples. For example, in the process of image transmission, the sequence may change, resulting in partial alignment between samples from different views. Therefore, achieving full alignment between views has garnered significant attention. Some end-to-end alignment and contrastive learning-based methods have been proposed; however, these methods still encounter challenges in mea-suring the similarity between high-dimensional samples and distinguishing among high-dimensional sample categories. In this paper, we propose a novel model, termed A Joint Facili-tation Alignment Model Based on Hyperbolic Non-Euclidean Mapping Distance and Contrastive Learning(HC-JFAM), to tackle them. Specifically, a two-layer encoder-decoder with an optimizer is explored to obtain better low-dimensional features. Furthermore, a new Hyperbolic Non-Euclidean Map-ping Distance (HNEMD) is designed to achieve cross-view sample alignment. Finally, the optimization function is re-designed to better discriminate between positive and negative pairs, and a new feature is obtained through weighted fusion for better clustering. Experimental results show that our pro-posed method improves the accuracy of sample alignment between views, and prove the effectiveness of HC-JFAM.