Accurate forecasting of electric vehicle (EV) charging demand is critical for intelligent energy management and sustainable urban mobility. However, existing methods often fall short in modeling the multimodal complexity of urban systems, where dynamic pricing interventions, exogenous signals, and latent functional semantics interact in non-trivial ways. In this paper, we propose a novel explainable multimodal framework, MCKNet, that leverages Kolmogorov-Arnold Networks (KANs) to capture non-linear demand patterns while providing interpretable insights into cross-modal interactions. MCKNet consists of three key components: a temporal-intervention branch that disentangles proactive pricing effects, a semantic prototyping module that aligns stations with latent functional roles, and a conflict-aware evidential fusion mechanism that models uncertainty and resolves modal inconsistencies. Experimental results on real-world urban datasets demonstrate that MCKNet outperforms state-of-the-art baselines in forecasting accuracy, robustness, and generalizability across cities. Furthermore, our framework provides transparent visual interpretations of spatial, temporal, and semantic factors that drive EV charging behavior.

Spiking neural P systems (SN PS) exemplify the third-generation of spiking neural networks (SNNs), which perform distributed and concurrent computations. However, SN PS rely solely on the spike as the singular signaling entity, ignoring the influence of other substances in the biological nervous system. Neuromodulators have the ability to exert their effects on the postsynaptic membrane, influencing synaptic plasticity and modifying the intensity of interneuronal connections. Motivated by this biological observation, we introduce a self-adaptation spiking neural P system with neuromodulators (SSNN PS). Specifically, neuromodulators generated by neurons are designed as resources consumed by rules in the postsynaptic membrane. The postsynaptic membrane, functioning as a new computational unit with three novel rules, possesses self-adapting weights regulated by neuromodulators and reflects the intensity of connections between neurons. Therefore, the SSNN PS enhance the control of the system over the computing process. In this work, we demonstrate the Turing universality of SSNN PS as both a number-generating and a number-accepting device. In addition, to verify the application capability, an SSNN PS for gender recognition of face images was constructed. The postsynaptic membrane self-adaptively updates its weights as it receives the feedback neuromodulators, which makes the recognition result more accurate. It achieves an accuracy of 91.71% on the UTKFace dataset and 87.83% on the FairFace dataset, and outperforms the other five comparative methods.

Multi-modal learning has become a transformative approach in recommendation systems, leveraging diverse data types—such as visual, textual, and audio signals—to construct rich and comprehensive user preference profiles. Despite significant progress, existing methods often struggle with key challenges, including imbalanced data utilization, diverse inter-modal correlations, and task-specific variability, which limit their ability to fully exploit inter-modal relationships. To address these issues, we propose KANM$^{2}$L (KAN Enhanced Multi-modal Learning for Recommendation), a novel framework that integrates the strengths of the Kolmogorov-Arnold Network (KAN) with multi-modal learning. Specifically, KANM$^{2}$L: (1) introduces a KAN-enhanced dilated attention mechanism to effectively capture high-dimensional, complex visual dependencies, enabling scalable and efficient processing of intricate datasets; (2) employs a multi-modal adversarial network to align and fuse features across modalities, ensuring seamless integration and improved recommendation accuracy; and (3) incorporates a rotational loss function to stabilize and refine visual feature embeddings, leveraging historical interaction data for more consistent performance. Extensive experiments on real-world datasets demonstrate that KANM$^{2}$L achieves state-of-the-art performance, with improvements of up to 11.7% over existing methods. These findings underscore the potential of KANM$^{2}$L to advance the field of multi-modal recommendation systems by overcoming critical limitations and delivering robust, scalable performance across diverse recommendation tasks.

Accurate electroencephalogram(EEG) annotation is essential for seizure detection but costly and error-prone, which can affect subsequent tasks. Moreover, the brain is a non-Euclidean topological structure, which contains spatial information for seizure detection. Based on these, this paper proposes a semi- supervised model based on Gaussian mixture variational autoencoder with graph representation, named GGMVAE. Firstly, for unlabeled EEG signals, we construct an adjacency matrix via Pearson correlation between channels. Then, matrix and EEG features are fed into a Gaussian Mixture VAE to learn the temporal and spatiall features through unsupervised training. Finally, the pre-trained encoder extracts low-dimensional features from partially labeled data for classification. The method is evaluated on the epilepsy dataset at the University of Helsinki, achieving the accuracy of 97.27 %, precision of 96.17 %, recall of 98.41 %, and F1-score of 97.28 %. The results indicate that this semi-supervised model can effectively learn the temporal and spatial features of EEG signals, improving the performance of seizure detection..

In time series analysis, disentangling long-term trends and seasonal patterns is crucial for capturing multi-scale temporal structures and improving both interpretability and forecasting accuracy. Recently, 2D modeling techniques have been incorporated into multivariate forecasting frameworks to better exploit periodic patterns. However, conventional decomposition methods often rely on simplistic moving averages that obscure critical patterns, while 2D modeling may entangle global trends with local variations and fail to normalize seasonal amplitudes-ultimately impairing both interpretability and forecast accuracy. To overcome these limitations, we propose XDNet (Exponential-Dimensional Network), a principled forecasting framework that explicitly disentangles trend and seasonal dynamics. At its core lies the Exponentially Weighted Decomposition (XWD), which applies decaying weights to past observations to preserve the integrity of long-term trends while adaptively normalizing seasonal fluctuations. The trend component is modeled using Temporal Kolmogorov-Arnold Networks (KAN) to capture intricate nonlinear dynamics, while the seasonal component is processed through a refined Inception-based module that robustly extracts fine-grained periodic dependencies. Extensive experiments on multiple benchmark datasets demonstrate that XDNet achieves state-of-the-art forecasting performance, delivering up to a 2.79% improvement in average accuracy over leading baselines, particularly in long-horizon prediction tasks.

Graph contrastive learning has demonstrated its effectiveness in time series modeling. However, existing approaches often face significant challenges, including sensitivity to noise, incompleteness in data, and the difficulty of balancing performance across long- and short-term forecasting tasks. To overcome these limitations, we propose a novel fusion-based framework, Contrastive Graph and Information Bottleneck Fusion Learning (FusionBC), tailored for robust multivariate time series forecasting. By integrating Contrastive Graph Learning with the Information Bottleneck principle, our method selectively filters out irrelevant or redundant information during the learning process, leading to refined, noise-resilient representations that enhance forecasting accuracy. Specifically, the FusionBC framework employs adaptive graph augmentation, intelligently dropping edges or nodes to optimize graph structures and reinforce robustness. Additionally, it combines scale-wise temporal convolution with dual-flow spatial attentive graph convolution, effectively capturing both inter-variable dynamics and multi-scale intra-variable dependencies. Extensive evaluations on real-world datasets show that FusionBC consistently outperforms state-of-the-art methods, achieving a 4.0% improvement in RSE on multivariate forecasting benchmarks. This improvement underscores FusionBC's ability to deliver enhanced, noise-resilient forecasting results that are robust across a range of time series scenarios.

Federated learning (FL), when integrated with blockchain, facilitates secure data sharing in autonomous driving applications. As vehicle-generated data becomes more granular and complex, the absence of data quality audits raises concerns about multiparty mistrust in trajectory prediction tasks. However, most of the existing research on trajectory prediction focuses on how to improve the model to enhance the prediction accuracy, and lacks the consideration of the privacy and security issues of data sharing in real-world scenarios. To address this, we propose an asynchronous FL data-sharing method, incorporating an interpretable reputation quantization mechanism based on graph convolutional networks. Data providers share data structures under differential privacy constraints, ensuring security while minimizing redundancy. We utilize deep reinforcement learning to classify vehicles by reputation level, optimizing FL aggregation efficiency. Experimental results show that the proposed scheme not only strengthens the security of trajectory prediction but also improves prediction accuracy.

Fine-grained urban flow inference is pivotal in alleviating traffic congestion and reducing detector deployment costs. It aims to infer fine-grained flow maps from coarse-grained traffic data. However, existing methods face challenges due to the highly complex nature of spatial modeling for urban flow patterns and the distinctive impact of external factors such as temperature and weather. To address these issues, this paper proposes a Simplified Multi-Factor spatial modeling framework (SimMF) to enhance the accuracy of fine-grained flow inference while optimizing inference complexity. SimMF incorporates a dual-path architecture for short-range modeling, combining multi-scale convolutions and frequency-domain processing via FFT to capture cross-scale spatial correlations and heterogeneity. For long-range dependencies, SimMF employs enhanced bottleneck attention with linear complexity, effectively modeling intricate spatial relationships. Additionally, SimMF adopts a view-aware learnable approach to represent external factors, enabling each factor to generate distinctive feature maps and capture its unique characteristics. Experimental results on two urban datasets demonstrate that SimMF outperforms existing methods, achieving superior inference accuracy while maintaining computational efficiency with significantly improved computational efficiency.

Urban region profiling is essential for forecasting and decision-making in dynamic and noisy urban environments. However, existing approaches struggle with adversarial attacks, data incompleteness, and security vulnerabilities, which undermine predictive accuracy and reliability. This paper introduces Enhanced Urban Region Profiling with Adversarial Self-Supervised Learning (EUPAS), a robust framework that integrates adversarial contrastive learning with self-supervised and supervised objectives. To fortify resilience against adversarial attacks and noisy data, we introduce perturbation augmentation, a trickster generator, and a deviation copy generator, which collectively enhance the robustness of learned embeddings. EUPAS significantly outperforms state-of-the-art models in forecasting tasks, including crime prediction, check-in prediction, and land use classification, achieving up to 12.2% improvement in forecasting performance. Additionally, our model demonstrates superior resilience against transfer-based black-box and white-box attacks compared to baseline models. By addressing key security challenges in data-driven urban modeling, EUPAS provides a scalable and adversarially robust solution for smart city applications.

Cell type annotation refers to the process of categorizing and labeling cells to identify their specific cell types, which is crucial for understanding cell functions and biological processes. Although many methods have been developed for automated cell type annotation, they often encounter challenges such as batch effects due to variations in data distribution across platforms and species, thereby compromising their performance. To address batch effects, in this study, a pre-trained domain adaptation model based on structural similarity, named pscAdapt, is proposed for cell type annotation. Specifically, a pre-trained strategy is employed to initialize model parameters to learn the data distribution of source domain. This strategy is also combined with an adversarial learning strategy to train the domain adaptation network for achieving domain level alignment and reducing domain discrepancy. Furthermore, to better distinguish different types of cells, a structural similarity loss is designed, aiming to shorten distances between cells of the same type and increase distances between cells of different types in feature space, thus achieving cell level alignment and enhancing the discriminability of cell types. Comprehensive experiments were conducted on simulated datasets, cross-platforms datasets and cross-species datasets to validate the effectiveness of pscAdapt, results of which demonstrate that pscAdapt outperforms several popular cell type annotation methods.

Spatial transcriptomic techniques can be used to obtain transcriptome data from different locations in tissues. The identification of spatial domains is a key task in the analysis of spatial transcriptomic data. Therefore, we propose a multi-view contrastive learning framework named MVCLST for identifying spatial domains. First, MVCLST introduces pathway information on the basis of spatial transcriptomic data to consider the functional correlation between spots. In order to better mine the underlying biological features, MVCLST constructs three biological networks from three biological perspectives: spatial distribution, similarity of gene expression and functional correlation. Secondly, MVCLST introduces a multi-view contrastive learning method, which fully considers the contrastive relationship between multiple views to improve the accuracy and reliability of feature extraction. Then, in order to obtain a more abundant feature representation, the adaptive attention mechanism is used to integrate the common features and specific features. Finally, we compare MVCLST with five spatial transcriptomic methods to verify the accuracy of MVCLST for spatial domains identification. The experimental results show that MVCLST is better than the other five methods.

Analysis of epistatic interaction is an important means to study the pathogenesis of complex diseases in genome-wide association studies (GWAS). Epistatic interaction detection aims to identify the ideal combination among single nucleotide polymorphisms (SNPs) and determine whether this combination is significantly associated with complex diseases. However, they suffer from certain limitations, such as low detection power and long execution times. Therefore, this paper proposes a multi-population ant colony optimization algorithm with an adaptive heuristic strategy (MPACO-AHS). MPACO-AHS is a framework based on multi-population approaches, where multiple populations are employed to detect epistatic interactions, helping to avoid the data bias inherent in a single population. Moreover, to guide the search direction of each population, an adaptive heuristic strategy is introduced, allowing the algorithm to focus on areas more likely to contain epistatic interactions, thereby improving the accuracy of the results. Comprehensive experiments are conducted on simulated datasets. The results demonstrate that MPACO-AHS outperforms existing algorithms by overcoming the challenges in detecting epistatic interactions in GWAS.

Single nucleotide polymorphism (SNPs) data have become abundant thanks to the quick advancement of high-throughput sequencing technology, which provides convenience for genome-wide association studies. Single SNPs have been proven to be the cause of some diseases, and the emergence of complex diseases is often thought to be the result of the interaction of multiple SNPs. However, the possible interaction of millions of SNPs imposes a heavy computational burden for uncovering complex disease mechanisms. The existing SNP-SNP interaction detection algorithms frequently have flaws including high computation complexity and poor optimization effectiveness. In this study, a particle swarm optimization algorithm based on multi-population mutual learning (PSOMPML) is proposed to detect SNP-SNP interactions. In this algorithm, the mutual learning strategy is introduced to deal with different particles in different sub-populations to facilitate knowledge exchange. In addition, the elite preservation mechanism is incorporated into PSOMPML, to better preserve the good SNPs in the elite particles. The promising region local search strategy searches the optimal solution along the target solution and its near space to increase the convergence speed of the proposed algorithm. Experiments on simulated data sets and real data also demonstrate the effectiveness of the proposed algorithm.

The problem of traffic forecasting has received much attention as a central part of intelligent transportation systems. In recent years, many different models have been proposed to improve the performance of traffic forecasting. However, there are some problems with these models: they only focus on the dependencies between nodes and ignore the dependencies between edges; the highly dynamic spatial dependencies of traffic networks in time are not fully considered. In this paper, we propose a multi-scale dynamic spatial-temporal graph convolution network with edge feature embedding(MDSTGCN). In the spatial dimension, we construct the dynamic adjacency matrix and the hypergraph. Capturing spatial correlations using diffusion convolution. In the temporal dimension, we design a multi-scale temporal convolution module to capture the temporal dynamics of traffic data at different scales. We conducted experiments on four real datasets and the results show that our model outperforms the baseline models.

Recently, contrastive learning has shown promising results for representing graphs. Despite their success, several key issues have not been well addressed in existing studies: 1) Data noise and incompleteness are inevitable in the graph signal due to various factors. 2) Improper augmentation strategies may have negative effects on views construction and graph representation. In this study, we propose a novel joint contrastive learning model for graph representation named MVJCL. Specifically, a set of views were constructed with topology-level and node-level augmentation strategies. For each view, we execute two-layers GCNs to learn the node embeddings. Then, we propose a positive-negative-positive (pnp) contrastive learning task, which performs contrastive learning between the negative view and each positive view, so as alleviate the noise of supervision signal and exploit the most critical information. Extensive experiments on five real-world datasets demonstrate the effectiveness of MVJCL, where the maximum improvement can reach to 4.63%.

As the urbanization accelerates, traffic congestion and accidents problems are more serious. Accurate traffic flow forecasting is essential for urban traffic management and optimization. Recently, graph neural networks (GNNs) based forecasting methods have obtained superior results. However, there are still two challenges: 1) The majority of traffic flow forecasting methods are rooted in supervised learning and lack considering data noise, while fewer studies focusing on contrastive learning. 2) Urban traffic networks exhibit not only local spatial dependencies but also global spatial correlations. To tackle these challenges, we propose a novel Urban Spatial-Temporal Graph Contrastive Learning framework(USTGCL). Specifically, we first perform augmentation on the traffic spatial-temporal data through topology and feature-level strategies to learn local and global information. Following a simple encoder, we apply contrastive learning auxiliary task in the high-level representations to jointly learn similarities and differences within the traffic data. Extensive experimental results demonstrate that our model outperforms the state-of-the-art baselines on two urban datasets, achieving an improvement of approximately 4.67%.

Graph contrastive learning under a single augmentation often fails to fully capture the properties of the graph. In order to address this limitation, it is beneficial to utilize hybrid augmentations, such as edge and node-level augmentations, as they enable the model to learn more robust semantic commonalities. However the combination of multiple augmentations often faces the challenge of complex parameter tuning, with augmentation rate being the most crucial. The magnitude of the augmentation rate directly affects the semantics of the contrastive views and is therefore crucial for the performance of the model. Although automated data augmentation methods have emerged, they either require labels or struggle to overcome the challenge of determining the appropriate augmentation rate, necessitating extensive evaluation and trial-and-error. To address these challenges, we propose a new framework (JTAGCL) that employs joint training for hybrid automated augmentations in graph contrastive learning. By doing so, it not only achieves automated optimization of augmentations without relying on labels but also adapts the augmentation rate in an adaptive manner. Furthermore, we design a sampling strategy to further alleviate the computational burden associated with the combination of two automated augmentations.

Traffic prediction is the cornerstone for enabling urban intelligent transportation systems. Robust and efficient prediction methods contribute to the city resources pre-allocation to manage traffic routine and relieve traffic congestion. Traffic prediction is a typical spatial-temporal prediction problem. However, the complex non-linear spatial-temporal dependencies make the problem challenging, especially for long-term prediction. In this paper, we propose a transformer-based traffic prediction framework (MPFormer) to effectively capture both long and short-term spatial as well as temporal dependencies to enhance the accuracy of traffic prediction. Specially, we introduce a temporal attention module that specifically focuses on capturing temporal dependencies, allowing the model to learn the dynamic patterns of traffic over time. Furthermore, we have designed two feature augmentation modules to emphasize local and global spatial dependencies from different perspectives, considering both short and long-range views. Moreover, multi-head attention mechanism and stacked layers are used for spatial-temporal feature fusion. Experiments on two real traffic datasets demonstrate the effectiveness of the MPFormer algorithm.

Although the density peak clustering (DPC) algorithm can effectively distribute samples and quickly identify noise points, it lacks adaptability and cannot consider the local data structure. In addition, clustering algorithms generally suffer from high time complexity. Prior research suggests that clustering algorithms grounded in P systems can mitigate time complexity concerns. Within the realm of membrane systems (P systems), spiking neural P systems (SN P systems), inspired by biological nervous systems, are third-generation neural networks that possess intricate structures and offer substantial parallelism advantages. Thus, this study first improved the DPC by introducing the maximum nearest neighbor distance and K-nearest neighbors (KNN). Moreover, a method based on delayed spiking neural P systems (DSN P systems) was proposed to improve the performance of the algorithm. Subsequently, the DSNP-ANDPC algorithm was proposed. The effectiveness of DSNP-ANDPC was evaluated through comprehensive evaluations across four synthetic datasets and 10 real-world datasets. The proposed method outperformed the other comparison methods in most cases.

Feature selection is a critical component of data mining and has garnered significant attention in recent years. However, feature selection methods based on information entropy often introduce complex mutual information forms to measure features, leading to increased redundancy and potential errors. To address this issue, we propose FSCME, a feature selection method combining Copula correlation (Ccor) and the maximum information coefficient (MIC) by entropy weights. The FSCME takes into consideration the relevance between features and labels, as well as the redundancy among candidate features and selected features. Therefore, the FSCME utilizes Ccor to measure the redundancy between features, while also estimating the relevance between features and labels. Meanwhile, the FSCME employs MIC to enhance the credibility of the correlation between features and labels. Moreover, this study employs the Entropy Weight Method (EWM) to evaluate and assign weights to the Ccor and MIC. The experimental results demonstrate that FSCME yields a more effective feature subset for subsequent clustering processes, significantly improving the classification performance compared to the other six feature selection methods.

The inherent complexities of single-cell RNA-seq data (scRNA-seq), such as high dimensionality, low signal-to-noise ratio, cellular heterogeneity, and imbalanced distribution of subcellular types, pose significant challenges when conducting cell type analysis. To address these obstacles, employing appropriate data preprocessing techniques in the single-cell clustering process is crucial, and spectral clustering is particularly effective due to its robustness to noise and outliers. Therefore, this paper presents a novel spectral clustering algorithm based on multi-similarity learning method (MSSC). However, utilizing a pairwise strategy to assess the similarity between two data points in conventional spectral clustering results in an insufficient representation of the intricate relationships within the dataset. In light of this issue, the proposed algorithm employs two similarity measurement methods, namely Euclidean distance and Spearman rank correlation coefficient, to obtain similarity matrices. These matrices are then fused for use in spectral clustering. Additionally, prior to performing spectral clustering, the scRNA-seq data is preprocessed using the Sigmoid kernel similarity method and normalization techniques. As a consequence, our method yields a more extensive and intricate dataset similarity information, thereby enhancing the performance of spectral clustering. Finally, the Gaussian mixture model (GMM) is used for clustering. In most cases, experiments validated that the MSSC method outperforms the other four clustering methods on seven benchmark scRNA-seq datasets.

In the genome-wide association study, the interactions of single nucleotide polymorphisms (SNPs) play an important role in revealing the genetic mechanism of complex diseases, and such interaction is called epistasis or epistatic interactions. In recent years, swarm intelligence methods have been widely used to detect epistatic interactions because they can effectively deal with global optimization problems. In this study, we propose a crow search algorithm based on information interaction (FICSA) to detect epistatic interactions. FICSA combines particle swarm optimization (PSO) and crow search algorithm (CSA) to balance the exploration and exploitation in the search process, which can effectively improve the ability of the algorithm to detect epistatic interactions. In addition, opposition-based learning strategy and adaptive parameters are used to further improve the performance of the algorithm. We compare FICSA with seven other epistasis detection algorithms using both simulated datasets and a real-life age-related macular degeneration (AMD) dataset. The results on simulated datasets show that FICSA has better detection power, while the results on the real dataset demonstrate the effectiveness of the proposed algorithm.

Urban region embedding is an important and yet highly challenging issue due to the complexity and constantly changing nature of urban data. To address the challenges, we propose a Region-Wise Multi-View Representation Learning (ROMER) to capture multi-view dependencies and learn expressive representations of urban regions without the constraints of rigid neighbourhood region conditions. Our model focuses on learn urban region representation from multi-source urban data. First, we capture the multi-view correlations from mobility flow patterns, POI semantics and check-in dynamics. Then, we adopt global graph attention networks to learn similarity of any two vertices in graphs. To comprehensively consider and share features of multiple views, a two-stage fusion module is further proposed to learn weights with external attention to fuse multi-view embeddings. Extensive experiments for two downstream tasks on real-world datasets demonstrate that our model outperforms state-of-the-art methods by up to 17% improvement.

Robust and accurate traffic forecasting is a key issue in intelligent transportation systems. Existing studies usually employ pre-defined spatial graph or learned fixed adjacency graph and design models to capture spatial and temporal features. However, pre-defined or fixed graph can not accurately model the complex hidden structure. Moreover, few solutions are satisfied with both long and short-term prediction tasks. In this paper, we propose a novel dual-stage graph convolution network based on graph learning (DSGCN) to address these challenges. To equip the graph convolution network with a flexible and practical graph structure, DSGCN designs a graph learning module to model the varying relations among nodes in the road network. In particular, we first provide a hierarchical graph structure cooperated with the dilated convolution to capture the temporal dependencies. Second, a dual-stage graph convolution layer is proposed to capture the complex spatial dependencies. Experiments on two real-world datasets demonstrate that DSGCN outperforms the state-of-the-art baselines, especially for long-term traffic prediction.

In this study, we proposed a novel method called the graph capsule convolutional network (GCCN) to predict the progression from mild cognitive impairment to dementia and identify its pathogenesis. First, we proposed a novel risk gene discovery component to indirectly target genes with higher interactions with others. These risk genes and brain regions were collected as nodes to construct heterogeneous pathogenic information association graphs. Second, the graph capsules were established by projecting heterogeneous pathogenic information into a set of disentangled latent components. The orientation and length of capsules are representations of the format and intensity of pathogenic information. Third, graph capsule convolution network was used to model the information flows among pathogenic factors and elaborates the convergence of primary capsules to advanced capsules. The advanced capsule is a concept that organizes pathogenic information based on its consistency, and the synergistic effects of advanced capsules directed the development of the disease. Finally, discriminative pathogenic information flows were captured by a straightforward built-in interpretation mechanism, i.e., the dynamic routing mechanism, and applied to the identification of pathogenesis. GCCN has been experimentally shown to be significantly advanced on public datasets. Further experiments have shown that the pathogenic factors identified by GCCN are evidential and closely related to progressive mild cognitive impairment.

It is well known that the study of the lncRNA-disease associations (LDAs) is of great value for the diagnosis and cure of many complex diseases. However, exploring unknown LDAs is extremely difficult and expensive. Therefore, it is essential to find a more accurate and effective calculation method to predict the potential LDAs. However, most previous studies focused on designing complex similarity-based methods to predict the potential interaction between lncRNAs and diseases. In this research, combining the three biological networks of lncRNA-disease, miRNA-lncRNA and miRNA-disease, a new computing model based on heterogeneous networks and stacked autoencoder (SAE) is proposed, called HSAELDA. Then, the SAE is used to extract the comprehensive features of the lncRNA-disease pairs, the LightGBM classifier is used for training. At the same time, five-fold cross-validation (CV) is used to compare our model with some existing prediction methods. The final comparison results showed HSAELDA obtained the highest AUC value of 0.978. In conclusion, the overall prediction performance of HSAELDA has been greatly improved compared to the state-of-art models. Experimental results and case study results show that HSAELDA is an effective method for predicting potential LDAs.

The pathogenesis of many human diseases is unclear, but many studies have shown that lncRNAs are deeply involved in the development of diseases. However, the exploration of lncRNA-disease associations in the laboratory requires a lot of time and financial resources, and computational-based methods have obvious advantages and become a promising research direction. But few experiments consider the relationship between other biological factors and lncRNAs and diseases. In this paper, a novel lncRNA-disease association prediction method, MANBNR, is proposed. MiRNAs are introduced by MANBNR to construct a lncRNA-miRNA-disease multilayer association network (MAN). The main innovation of MANBNR is to mine potential lncRNA-disease association information based on miRNA information. For any lncRNA-disease pair, lncRNA-associated miRNAs and disease-associated miRNAs are sorted into two sets, respectively. The association of this lncRNA-disease pair is judged by comparing the number of miRNAs shared in the two sets. This solves the problem that the known lncRNA-disease association matrices are too sparse. Finally, the bipartite network recommendation (BNR) algorithm was used to accurately predict potential lncRNA-disease association. The performance of MANBNR is better than that of many advanced methods at present. Case studies of breast cancer and lung cancer further demonstrate that MANBNR is an effective and reliable method for LDAs prediction.

Predicting disease-related long non-coding RNAs (lncRNAs) can help reveal the genetic mechanisms of complex diseases. Accurately identifying disease-associated lncRNAs is crucial for human diagnosis and therapeutics of complex diseases. However, most computational models ignore the noise of the data and the interference of redundant information. In this study, we build heterogeneous networks by integrating three different data sources of lncRNAs, miRNAs and diseases, and then propose an efficient computational model called MHILDA. MHILDA selects the most helpful features to train the model by Lasso's feature extraction. MHILDA is evaluated by five-fold cross-validation and performs well both on the benchmark dataset and on the independent test set. To further evaluate the performance of MHILDA, two types of case studies are implemented. The experimental results show that MHILDA can predict lncRNAs for unknown diseases.

Traffic forecasting is essential for transportation services such as traffic control and route planning. However, accurate traffic prediction is challenging due to complex characteristics of traffic data. Existing solutions may not adequately capture dynamic and nonlinear spatial-temporal correlations in traffic network. In this paper, we propose a novel Multi-Hierarchical Spatial-Temporal Graph Convolutional Networks (MH-GCN) to solve traffic flow forecasting problem. It adopts an attention-based encoder-decoder structure. Firstly, MH-GCN uses a spatial-temporal attention mechanism in encoder to model dynamic spatial and nonlinear temporal correlations. Then, a transformer attention layer is positioned between encoder and decoder, which is used to model the correlation of historical and future time. Finally, the decoder utilizes Convolution Group, Pooling Group, and Dilation Group to extract different hierarchical of characteristics from the already modeled features, and then the fused results are used for predicting future traffic conditions. Experiments on two real traffic datasets demonstrate that the proposed MH-GCN obtains improvements over the state-of-the-art baselines.

Robust and accurate traffic forecasting is a hot issue in Intelligent Transportation Systems (ITS). It is helpful in alleviating traffic congestion, which improves the efficiency of urban road traffic. The highly non-linear and dynamic spatial-temporal correlations propose challenges for timely accurate traffic forecasting, especially long-term forecasting. Existing studies have considered these problems and proposed solutions. However, few studies are satisfied with both long- and short-term prediction tasks. In this paper, we propose a novel Long- and Short-term Transformer networks (LSTN) to address these challenges. LSTN employs the Recurrent Neural Network (RNN) and multi-head attention mechanism to discover long-term patterns and model long-term bidirectional dependencies for time series. To solve the problem of scale insensitive problem of the neural network model, we further use the traditional autoregressive model. Experiments on two real-world datasets PeMS03 and PeMS08 demonstrate that LSTN outperforms the state-of-the-art baselines, especially for long-term traffic flow forecasting.

Traffic prediction is a critical component of intel-ligent transportation systems. However, highly non-linear and dynamical spatial-temporal correlations propose challenges for traffic prediction, especially long-term prediction. We propose a spatial-temporal channel-attention based graph convolutional network (STCAGCN) to improve the accuracy of both long-term and short-term traffic flow prediction. Firstly we design an attention mechanism to learn complex temporal and spatial correlations. Then we develop the stacked spatial-temporal convo-lution layer to model complex temporal and spatial correlations. Each spatial-temporal convolution layer is composed of a gated time convolution network and a graph convolution network. We develop a gated time convolution network to model non-linear temporal correlations, which process long sequences through stacked dilated convolution. Moreover, the graph convolution network exploits the hidden spatial correlations via learning self-adaptive adjacency matrix. Experiment results on real-world datasets demonstrate that the proposed STCAGCN model obtains improvements over the state-of-the-art, especially for long-term traffic flow prediction.

Recently, unsupervised learning is proposed to avoid the performance degrading caused by synthesized paired computed tomography (CT) images. However, existing unsupervised methods for metal artifact reduction (MAR) only use features in image space, which is not enough to restore regions heavily corrupted by metal artifacts. Besides, they lack the distinction and selection for effective features. To address these issues, we propose an attention-embedded decomposed network to reducing metal artifacts in both image space and sinogram space with unpaired images. Specifically, combining with the CT images prior, we decompose the artifact-affected images to artifact images and content images. Besides, normal convolutions are embedded with attention design in pixel-wise and channel-wise to strengthen the representational capacity. Extensive experiments show notable improvements on both synthesized data and clinical data.

Distant supervised relation extraction has recently attracted researchers attention in the knowledge graph.However, the current feature encoding model of sentences can not fully represent the features in sentences, which poses a challenge.To solve this problem, we propose a two-dimensional feature encodingbased sentence-level attention model for relation extraction.In this model, we first employ bidirectional long short-term memory networks(BiLSTM) to capture the temporal dependency of the words in the sentence.Then we employ multi-dilated convolution to obtain the higher-level semantic units hidden in the sentence.Afterwards, we combine the above two-dimensional features to embed the encoding of sentences, which is expected to enhance the model's ability to express sentence features.Finally we build sentence-level attention to complete the relation extraction task.Compared with other excellent methods, the proposed approach provides a significant performance improvement.

Target tracking is a well studied topic in wireless sensor networks. However, uncertainty existed in sensor networks presents new challenges for it. Besides the energy conservation of networks, target tracking has to deal with different kinds of uncertainty, such as the impreciseness of positioning systems, environment noise and limited sensitivity of sensors. In this paper, we study the problem of target tracking under uncertainty in wireless sensor networks. We first investigate the uncertainty existed in sensor networks and propose a series of general models. Then, we introduce the problem of probabilistic k-nearest neighbors (PkNN) and provide an efficient tracking algorithm based on PkNN retrieval under the proposed models. Finally, a comprehensive set of simulations are presented. We conclude that the proposed tracking algorithm yields excellent tracking performance in wireless sensor networks.

Aggregations of sensed data are very important for users to get summary information about monitored area in applications of wireless sensor networks (WSNs). As the approximate aggregation results are enough for users to perform analysis and make decisions, many approximate aggregation algorithms are proposed for WSNs. However, most of the algorithms have fixed error bounds and cannot meet arbitrary precision requirement, the uniform sampling based algorithm which can reach arbitrary precision is just suitable for the static networks. Considering the dynamic property of WSNs, in this paper, we propose an approximate aggregation algorithm based on Bernoulli sampling to satisfy arbitrary precision requirement. Besides, two adaptive algorithms are also proposed, one is for adapting the sample with varying of precision requirement, the other is for adapting the sample with varying of sensed data. The theoretical analysis and experiment results show that the proposed algorithms have high performance in terms of accuracy and energy consumption.

Lifetime maximization is an important factor in the design of sensor networks for object tracking applications. Some techniques of node scheduling have been proposed to reduce energy consumption. By exploiting the redundancy of network coverage, they turn off unnecessary nodes, or make nodes work in turn, which require high nodes density or sacrificing tracking quality. We present TPSS, a two-phase sleep scheduling protocol, which divides the whole tracking procedure into two phases and assigns different scheduling policies at each phase. To balance energy savings and tracking quality, we further optimize the node scheduling protocol in terms of network coverage and nodes state prediction. We evaluate our method in an indoor environment with 36 sensor nodes. Comparing with the existing methods, all the experimental results show that our method has a high performance in term of energy savings and tracking quality.