Search

Matrix computing is the core component of machine learning and artificial intelligence. Fast matrix computations can facilitate many large-scale computational projects greatly. Basic linear algebra subprograms (BLAS) are proposed, which classify different matrices and provide a standardized interface. Currently, the most commonly used heterogeneous computing platforms are central processing unit (CPU) and graphics processing unit (GPU). At present, BLAS has been implemented on both CPU and GPU. However, due to the different characteristics of algorithms and hardware, a particular matrix method should be designed for a particular processor. It is important to choose the right processor for a particular matrix computation. This paper first briefly reviews the BLAS, and then introduces architecture and optimization methods of CPU and GPU. The effect of different subroutines in BLAS is studied through experiments. Finally, we discuss the reasons and the processor selection scheme of matrix computations. [ABSTRACT FROM AUTHOR]

Deep learning has achieved great successes in conventional computer vision tasks. In this paper, we exploit deep learning techniques to address the hyperspectral image classification problem. In contrast to conventional computer vision tasks that only examine the spatial context, our proposed method can exploit both spatial context and spectral correlation to enhance hyperspectral image classification. In particular, we advocate four new deep learning models, namely, 2-D convolutional neural network (2-D-CNN), 3-D-CNN, recurrent 2-D CNN (R-2-D-CNN), and recurrent 3-D-CNN (R-3-D-CNN) for hyperspectral image classification. We conducted rigorous experiments based on six publicly available data sets. Through a comparative evaluation with other state-of-the-art methods, our experimental results confirm the superiority of the proposed deep learning models, especially the R-3-D-CNN and the R-2-D-CNN deep learning models. [ABSTRACT FROM AUTHOR]

Recently, due to the advancement of acquisition techniques, visual tensor data have been accumulated in a great variety of engineering fields, e.g., biometrics, neuroscience, surveillance and remote sensing. How to analyze and learn with such tensor objects thus becomes an important and growing interest in machine learning community. In this paper, we propose a block linear discriminant analysis (BLDA) algorithm to extract features for visual tensor objects such as multichannel/hyperspectral face images or human gait videos. Taking the inherent characteristic of such tensor data into account, we unfold tensor objects according to their spatial information and frequency/time information, and represent them in a block matrix form. As a result, the block form between-class and within-class scatter matrices are constructed, and a related block eigen-decomposition is solved to extract features for classification. Comprehensive experiments have been carried out to test the effectiveness of the proposed method, and the results show that BLDA outperforms existing algorithms like DATER, 2DLDA, GTDA, UMLDA, STDA and MPCA for visual tensor object analysis. [ABSTRACT FROM AUTHOR]

Time-stamped data are ubiquitous in our daily life, such as twitter data, academic papers and sensor data. Finding clusters and their evolutionary trends in time-stamped data sets are receiving increasing attention from researchers. Most existing methods, however, can only tackle the clustering problem of a data set without time-stamped information which is inherent in almost all the data objects. Actually, not only the performance can be improved by effectively incorporating the time-stamped information in the clustering process on most data sets, but also we can find the evolutionary trends of the clusters with time information. In this paper, we introduce an approach for clustering time-stamped data and discovering the evolutionary trends of the clusters by using Multiple Nonnegative Matrices Factorization (MNMF) with smooth constraint over time. To utilize time-stamped information in the clustering process, an extra object-time matrix is constructed in our proposed method. Then, we jointly factorize multiple feature matrices using smooth constraint to perform the object-time matrix to obtain the clusters and their evolutionary trends. Experimental results on real data sets demonstrate that our proposed approach outperforms the comparative algorithms with respect to Fscore, NMI or Entropy. [ABSTRACT FROM AUTHOR]

Most of kmeans-type clustering algorithms rely on only intra-cluster compactness, i.e. the dispersions of a cluster. Inter-cluster separation which is widely used in classification algorithms, however, is rarely considered in a clustering process. In this paper, we present a new discriminative subspace kmeans-type clustering algorithm (DSKmeans), which integrates the intra-cluster compactness and the inter-cluster separation simultaneously. Different to traditional weighting kmeans-type algorithms, a 3-order tensor is constructed to evaluate the importance of different features in order to integrate the aforementioned two types of information. First, a new objective function for clustering is designed. To optimize the objective function, the corresponding updating rules for the algorithm are then derived analytically. The properties and performance of DSKmeans are investigated on several numerical and categorical data sets. Experimental results corroborate that our proposed algorithm outperforms the state-of-the-art kmeans-type clustering algorithms with respects to four metrics: Accuracy, RandIndex, Fscore and Normal Mutual Information(NMI). [ABSTRACT FROM AUTHOR]

Unsupervised hashing is an important approach for large-scale content-based remote sensing (RS) image retrieval. Existing unsupervised hashing methods usually utilize data clustering to generate pseudo labels as supervised signals. In the RS domain, images of each class cannot be accurately grouped into individual clusters because image features are not well characterized by the pre-trained models learned on a nature image dataset. As a result, to preserve the similarity of images sharing at least one class, intra-cluster and inter-cluster similarities need to be carefully learned. However, existing pseudo-labels are hard pseudo-labels represented by scalar values, which cannot well reflect the semantic distance between inter-cluster images or the semantic distance between intra-cluster images. To address these problems, this paper proposes a soft-pseudo-label -based unsupervised deep hashing method for content-based RS image retrieval, called SPL-UDH. Soft pseudo-labels can accurately describe the global similarity between inter-cluster images by binarized vectors. Specifically, we design a deep auto-encoder network to learn soft pseudo-labels automatically and meanwhile to generate a local similarity matrix representing the proximity between intra-cluster images. Based on soft pseudo-labels and local similarity matrix, we propose a deep hashing network to simultaneously learn the inter-cluster similarity and the intra-cluster similarity between RS images. Moreover, we design a new objective function based on Bayesian theory so that the deep hashing network can be trained by jointly learning the soft pseudo-labels and the local similarity matrix. Extensive experiments on public RS image retrieval datasets demonstrate that SPL-UDH outperforms various state-of-the-art unsupervised hashing methods. [ABSTRACT FROM AUTHOR]

Abstract: This paper presents a Crotch Ensemble classification model for high dimensional data. A Crotch Ensemble is obtained from a decision cluster tree built by calling a clustering algorithm recursively. A crotch is an inner node of the tree together with its direct children. If the children of a crotch have more than one dominant class, the crotch is defined as a crotch predictor. Each crotch predictor constructs a classifier by itself. A Crotch Ensemble consists of a set of crotch predictors. When classifying a new object, a subset of crotch predictors is selected according to the distances between the object and the crotch predictors. A classification is made on the object as the class predicted by the crotch predictors with the maximum accumulative weights. The experimental results on both synthetic and real data have shown that the Crotch Ensemble model can get better classification results on high dimensional data than other classification methods. [Copyright &y& Elsevier]

Abstract: For high dimensional data a large portion of features are often not informative of the class of the objects. Random forest algorithms tend to use a simple random sampling of features in building their decision trees and consequently select many subspaces that contain few, if any, informative features. In this paper we propose a stratified sampling method to select the feature subspaces for random forests with high dimensional data. The key idea is to stratify features into two groups. One group will contain strong informative features and the other weak informative features. Then, for feature subspace selection, we randomly select features from each group proportionally. The advantage of stratified sampling is that we can ensure that each subspace contains enough informative features for classification in high dimensional data. Testing on both synthetic data and various real data sets in gene classification, image categorization and face recognition data sets consistently demonstrates the effectiveness of this new method. The performance is shown to better that of state-of-the-art algorithms including SVM, the four variants of random forests (RF, ERT, enrich-RF, and oblique-RF), and nearest neighbor (NN) algorithms. [Copyright &y& Elsevier]

Abstract: This paper proposes a new method to weight subspaces in feature groups and individual features for clustering high-dimensional data. In this method, the features of high-dimensional data are divided into feature groups, based on their natural characteristics. Two types of weights are introduced to the clustering process to simultaneously identify the importance of feature groups and individual features in each cluster. A new optimization model is given to define the optimization process and a new clustering algorithm FG-k-means is proposed to optimize the optimization model. The new algorithm is an extension to k-means by adding two additional steps to automatically calculate the two types of subspace weights. A new data generation method is presented to generate high-dimensional data with clusters in subspaces of both feature groups and individual features. Experimental results on synthetic and real-life data have shown that the FG-k-means algorithm significantly outperformed four k-means type algorithms, i.e., k-means, W-k-means, LAC and EWKM in almost all experiments. The new algorithm is robust to noise and missing values which commonly exist in high-dimensional data. [Copyright &y& Elsevier]

Abstract: Clustering is one of the important data mining tasks. Nested clusters or clusters of multi-density are very prevalent in data sets. In this paper, we develop a hierarchical clustering approach—a cluster tree to determine such cluster structure and understand hidden information present in data sets of nested clusters or clusters of multi-density. We embed the agglomerative k-means algorithm in the generation of cluster tree to detect such clusters. Experimental results on both synthetic data sets and real data sets are presented to illustrate the effectiveness of the proposed method. Compared with some existing clustering algorithms (DBSCAN, X-means, BIRCH, CURE, NBC, OPTICS, Neural Gas, Tree-SOM, EnDBSAN and LDBSCAN), our proposed cluster tree approach performs better than these methods. [Copyright &y& Elsevier]

Wind power is a new type of green energy. Though it is economical to access and gather such energy, effectively matching the energy with consumers' demand is difficult, because of the fluctuate, intermittent and chaotic nature of wind speed. Hence, multi-step wind speed prediction becomes an important research topic. In this paper, we propose a novel deep learning method, DyanmicNet, for the problem. DynamicNet follows an encoder–decoder framework. To capture the fluctuate, intermittent and chaotic nature of wind speed, it leverages a time-variant structure to build the decoder, which is different from conventional encoder–decoder methods. In addition, a new neural block (ST-GRU-ODE) is developed, which can model the wind speed in a continuous manner by using the neural ordinary differential equation (ODE). To enhance the prediction performance, a multi-step training procedure is also put forward. Comprehensive experiments have been conducted on two real-world datasets, where wind speed is recorded in the form of two orthogonal components namely U-Wind and V-Wind. Each component can be illustrated as wind speed images. Experimental results demonstrate the effectiveness and superiority of the proposed method over state-of-the-art techniques. [ABSTRACT FROM AUTHOR]

Satellite image sequence prediction is a crucial and challenging task. Previous studies leverage optical flow methods or existing deep learning methods on spatial–temporal sequence models for the task. However, they suffer from either oversimplified model assumptions or blurry predictions and sequential error accumulation issue, for a long-term forecast requirement. In this article, we propose a novel multiscale time conditional generative adversarial network (MSTCGAN). To address the sequential error accumulation issue, MSTCGAN adopts a parallel prediction framework to produce the future image sequences by a one-hot time condition input. In addition, a powerful multiscale generator is designed with the multihead axial attention, which helps to carefully preserve the fine-grained details for appearance consistency. Moreover, we develop a temporal discriminator to address the blurry issue and maintain the motion consistency in prediction. Extensive experiments have been conducted on the FengYun-4A satellite dataset, and the results demonstrate the effectiveness and superiority of the proposed method over state-of-the-art approaches. [ABSTRACT FROM AUTHOR]

Precipitation nowcasting is an important task, which can be used in numerous applications. The key challenge of the task lies in radar echo map prediction. Previous studies leverage the convolutional recurrent neural network (ConvRNN) to address the problem. However, the approaches are built upon mean square losses, and the results tend to have inaccurate appearances, shapes, and positions for predictions. To alleviate this problem, we explore the idea of adversarial regularization and systematically compare four types of generative adversarial networks (GANs), which are the combinations of GAN/Wasserstein GAN (WGAN) and its multiscale version. Extensive experiments on a real-world radar dataset and four typical meteorological examples are conducted. The results validate the effectiveness of adversarial regularization. The developed models show superior performances over the existing prediction approaches in the majority of circumstances. Moreover, we find that the WGAN regularization often delivers better results than the GAN regularization due to its robustness, and the multiscale WGAN, in general, performs the best among all the methods. To reproduce the results, we release the source code at https://github.com/luochuyao/MultiScaleGAN and the test system at http://39.97.217.145:80/. [ABSTRACT FROM AUTHOR]

Cross-modal hashing is an important tool for retrieving useful information from very-high-resolution (VHR) optical images and synthetic aperture radar (SAR) images. Dealing with the intermodal discrepancies, including both spatial–spectral and visual semantic aspects, between VHR and SAR images is extremely vital to generate high-quality common hash codes in the Hamming space. However, existing cross-modal hashing methods ignore the spatial–spectral discrepancy when representing VHR and SAR images. Moreover, existing methods employ derived supervised signals, such as pairwise training images, to implicitly guide hashing learning, which fails to effectively deal with the visual semantic discrepancy, i.e., cannot adequately preserve the intraclass similarity and interclass discrimination between VHR and SAR images. To address these drawbacks, this article proposes a multisensor fusion and explicit semantic preserving-based deep Hashing method, termed as MsEspH, which can effectively deal with the discrepancies. Specifically, we design a novel cross-modal hashing network to eliminate the spatial–spectral discrepancies by fusing extra multispectral images (MSIs), which are generated in real time by a generative adversarial network. Then, we propose an explicit semantic preserving-based objective function by analyzing the connection between classification and hash learning. The objective function can preserve the intraclass similarity and interclass discrimination with class labels directly. Moreover, we theoretically verify that hash learning and classification can be unified into a learning framework under certain conditions. To evaluate our method, we construct and release a large-scale VHR-SAR image dataset. Extensive experiments on the dataset demonstrate that our method outperforms various state-of-the-art cross-modal hashing methods. [ABSTRACT FROM AUTHOR]

Anomaly detection in multivariate time series is of critical importance in many real-world applications, such as system maintenance and Internet monitoring. In this article, we propose a novel unsupervised framework called SVD-AE to conduct anomaly detection in multivariate time series. The core idea is to fuse the strengths of both SVD and autoencoder to fully capture complex normal patterns in multivariate time series. An asymmetric autoencoder architecture is proposed, where two encoders are used to capture features in time and variable dimensions and a shared decoder is used to generate reconstructions based on latent representations from both dimensions. A new regularization based on singular value decomposition theory is designed to force each encoder to learn features in the corresponding axis with mathematical supports delivered. A specific loss component is further proposed to align Fourier coefficients of inputs and reconstructions. It can preserve details of original inputs, leading to enhanced feature learning capability of the model. Extensive experiments on three real world datasets demonstrate the proposed algorithm can achieve better performance on multivariate time series anomaly detection tasks under highly unbalanced scenarios compared with baseline algorithms. [ABSTRACT FROM AUTHOR]

Multi-label classification addresses the problem that each instance is associated with multiple labels simultaneously. In this paper, we propose a multi-label crotch ensemble (MLCE) model for multi-label classification, which takes label correlations into consideration. In MLCE, a multi-label cluster tree is first constructed. Then, we incorporate all multi-label crotch predictors of the tree into a classifier, where the multi-label crotch predictor is the crotch formed by an inner node of the tree and its children. Finally, a flexible weighted voting scheme is designed to produce the classification output. We perform experiments on 11 benchmark datasets. Experimental results clearly demonstrate the MLCE significantly outperforms six well-established multi-label classification approaches, in terms of the widely used evaluation metrics. [ABSTRACT FROM AUTHOR]

As a pixel-wise dense forecast task, video prediction is challenging due to its high computation complexity, dramatic future uncertainty, and extremely complicated spatial–temporal patterns. Many deep learning methods are proposed for the task, which bring up significant improvements. However, they focus on modeling short-term spatial–temporal dynamics and fail to sufficiently exploit long-term ones. As a result, the methods tend to deliver unsatisfactory performance for a long-term forecast requirement. In this article, we propose a novel unified memory network (UNIMEMnet) for long-term video prediction, which can effectively exploit long-term motion-appearance dynamics and unify the short-term spatial–temporal dynamics and long-term ones in an architecture. In the UNIMEMnet, a dual branch multi-scale memory module is carefully designed to extract and preserve long-term spatial–temporal patterns. In addition, a short-term spatial–temporal dynamics module and an alignment and fusion module are devised to capture and coordinate short-term motion-appearance dynamics with long-term ones from our designed memory module. Extensive experiments on five video prediction datasets from both synthetic and real-world scenarios are conducted, which validate the effectiveness and superiority of our proposed method UNIMEMnet over state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Network embedding aims to learn low-dimensional representations for nodes in social networks, which can serve many applications, such as node classification, link prediction and visualization. Most of network embedding methods focus on learning the representations solely from the topological structure. Recently, attributed network embedding, which utilizes both the topological structure and node content to jointly learn latent representations, becomes a hot topic. However, previous studies obtain the joint representations by directly concatenating the one from each aspect, which may lose the correlations between the topological structure and node content. In this paper, we propose a new attributed network embedding method, TLVANE, which can address the drawback by exploiting the deep variational autoencoders (VAEs). Particularly, a two-level VAE model is built, where the first-level accounts for the joint representations while the second for the embeddings of each aspect. Extensive experiments on three real-world datasets have been conducted, and the results demonstrate the superiority of the proposed method against state-of-the-art competitors. [ABSTRACT FROM AUTHOR]

A large set of training samples is a prerequisite to effectively learn deep neural networks for image classification. When the labeled samples are scarce, it often fails to produce a promising model. Data augmentation is a widely used technique to overcome the issue, which enlarges the training samples with label invariant transformations, e.g., rotation, flip and random crop etc. However, the diversity of images generated by standard data augmentation is quite limited and thus the final improvement on classification accuracy is not much, especially for fine-grained classification problem. In this paper, we propose a two-stage generative adversarial network, namely Fine-grained Conditional Adversarial Network (F-CGAN), which can produce class-dependent synthetic images with fine-grained details. Moreover, to leverage the synthetic images for fine-grained classification, we develop a multi-task learning classifier, which categorizes training images and synthetic images simultaneously. Experimental results on CUB Birds and Stanford Dogs data sets show that the proposed method indeed improves the classification accuracy. [ABSTRACT FROM AUTHOR]

Road information extraction based on aerial images is a critical task for many applications, and it has attracted considerable attention from researchers in the field of remote sensing. The problem is mainly composed of two subtasks, namely, road detection and centerline extraction. Most of the previous studies rely on multistage-based learning methods to solve the problem. However, these approaches may suffer from the well-known problem of propagation errors. In this paper, we propose a novel deep learning model, recurrent convolution neural network U-Net (RCNN-UNet), to tackle the aforementioned problem. Our proposed RCNN-UNet has three distinct advantages. First, the end-to-end deep learning scheme eliminates the propagation errors. Second, a carefully designed RCNN unit is leveraged to build our deep learning architecture, which can better exploit the spatial context and the rich low-level visual features. Thereby, it alleviates the detection problems caused by noises, occlusions, and complex backgrounds of roads. Third, as the tasks of road detection and centerline extraction are strongly correlated, a multitask learning scheme is designed so that two predictors can be simultaneously trained to improve both effectiveness and efficiency. Extensive experiments were carried out based on two publicly available benchmark data sets, and nine state-of-the-art baselines were used in a comparative evaluation. Our experimental results demonstrate the superiority of the proposed RCNN-UNet model for both the road detection and the centerline extraction tasks. [ABSTRACT FROM AUTHOR]

In wind speed prediction technologies, deep learning-based methods have achieved promising advantages. However, most existing methods focus on learning implicit knowledge in a data-driven manner but neglect some explicit knowledge from the physical theory of meteorological dynamics, failing to make stable and long-term predictions. In this paper, we explore introducing explicit physical knowledge into neural networks and propose Physical Equations Predictive Network (PEPNet) for multi-step wind speed predictions. In PEPNet, a new neural block called the Augmented Neural Barotropic Equations (ANBE) block is designed as its key component, which aims to capture the wind dynamics by combining barotropic primitive equations and deep neural networks. Specifically, the ANBE block adopts a two-branch structure to model wind dynamics, where one branch is physic-based and the other is data-driven-based. The physic-based branch constructs temporal partial derivatives of meteorological elements (including u-component wind, v-component wind, and geopotential height) in a new Neural Barotropic Equations Unit (NBEU). The NBEU is developed based on the barotropic primitive equations mode in numerical weather prediction (NWP). Besides, considering that the barotropic primitive mode is a crude assumption of atmospheric motion, another data-driven-based branch is developed in the ANBE block, which aims at capturing meteorological dynamics beyond barotropic primitive equations. Finally, the PEPNet follows a time-variant structure to enhance the model's capability to capture wind dynamics over time. To evaluate the predictive performance of PEPNet, we have conducted several experiments on two real-world datasets. Experimental results show that the proposed method outperforms the state-of-the-art techniques and achieve optimal performance. [ABSTRACT FROM AUTHOR]

Abstract Most of existing image categorizations assume that the given datasets have a good resolution and quality. However, the assumption is often violated in real applications. In this paper, we study the low-resolution (LR) image categorization. By utilizing labeled high-resolution (HR) images as auxiliary information, we formulate the problem as a heterogeneous domain adaptation problem and propose a Discriminative Joint Distribution Adaptation (DJDA) model to solve it. The DJDA model projects both LR and HR images into an intermediate subspace with a well-designed objective function, where the distance between classes is expected to be enlarged and the distribution divergence to be reduced. As a result, the discriminative knowledge for HR images can be transferred effectively to LR images. Experimental results demonstrate the proposed DJDA method produces significantly superior categorization accuracies against state-of-the-art competitors. [ABSTRACT FROM AUTHOR]

• We propose an interpretable partial differential network for SPLT-3D, where both horizontal and vertical interactions are modeled effectively. • We develop the theory of formulating 3D spatiotemporal dynamics with PDEs and design an effective strategy to apply it. • Experimental results show that our method can not only achieve better prediction performance, but also learn the correct PDEs. Neural network-based methods have been widely applied to spatiotemporal prediction tasks, such as video prediction and weather forecasting. However, most existing works are designed for prediction in 2D space, and 3D prediction has not been extensively studied. In this paper, we propose to leverage 3D partial differential equations (PDEs) for spatiotemporal prediction in 3D space, and further develop a novel 3D neural partial differential network. This is inspired by that 3D PDEs can model both horizontal and vertical information interactions by various partial derivatives. Moreover, they can also formulate physical knowledge by equations. To integrate 3D PDEs in neural networks, we first develop the theory of approximating 3D partial derivatives by 3D convolutions, and further present an effective strategy to utilize the theory in practice. Then based on the theory and strategy, we propose a novel 3D Neural Partial Differential Network for prediction, named NPDN-3D. Specifically, NPDN-3D consists of two pivotal modules: (1) a neural partial differential module for capturing low-order spatiotemporal dynamics. This module is the key for prediction, where the dynamics are formulated by commonly-used low-order 3D PDEs. (2) A residual module for capturing the remaining non-low-order dynamics. This module performs as an extensible plug-in to enhance the expressiveness of our model. Extensive experiments on two simulated datasets and two real datasets show that our method not only achieves better prediction but also learns the correct PDEs. [ABSTRACT FROM AUTHOR]

Regional wind speed prediction plays an important role in the development of wind power, which is usually recorded in the form of two orthogonal components, namely U-wind and V-wind. The regional wind speed has the characteristics of diverse variations, which are reflected in three aspects: (1) The spatially diverse variations of regional wind speed indicate that wind speed has different dynamic patterns at different positions; (2) The distinct variations between U-wind and V-wind denote that U-wind and V-wind at the same position exhibit different dynamic patterns; (3) The non-stationary variations of wind speed represent that the intermittent and chaotic nature of wind speed. In this paper, we propose a novel framework named Wind Dynamics Modeling Network (WDMNet) to model the diverse variations of regional wind speed and make accurate multi-step predictions. To jointly capture the spatially diverse variations and the distinct variations between U-wind and V-wind, WDMNet leverages a new neural block called Involution Gated Recurrent Unit Partial Differential Equation (Inv-GRU-PDE) as its key component. The block adopts involution to model spatially diverse variations and separately constructs hidden driven PDEs of U-wind and V-wind. The construction of PDEs in this block is achieved by a new Involution PDE (InvPDE) layers. Besides, a deep data-driven model is also introduced in Inv-GRU-PDE block as the complement to the constructed hidden PDEs for sufficiently modeling regional wind dynamics. Finally, to effectively capture the non-stationary variations of wind speed, WDMNet follows a time-variant structure for multi-step predictions. Comprehensive experiments have been conducted on two real-world datasets. Experimental results demonstrate the effectiveness and superiority of the proposed method over state-of-the-art techniques. [ABSTRACT FROM AUTHOR]

The real-world networking data may contain different types of attribute views and relational view. Hence, it is desirable to collectively use available attribute views and relational view in order to build effective learning models. We call this framework multi-attribute and relational learning. Collective classification is one of the popular approaches that can handle both attribute and relational information for network data. However, in collective classification only one type of attribute and relational view is involved and little attention is received for multi-attribute and relational learning. In this paper, we propose a new semi-supervised collective classification approach, called hypergraph regularized generative model (HRGM), for multi-attribute and relational learning. In the approach, a generative model based on the Probabilistic Latent Semantic Analysis (PLSA) method is developed to leverage attribute information, and a hypergraph regularizer is incorporated to effectively exploit higher-order relational information among the data samples. Experimental results on various data sets have demonstrated the effectiveness of the proposed HRGM, and revealed that our approach outperforms existing collective classification methods and multi-view classification methods in terms of accuracy. [ABSTRACT FROM AUTHOR]

• Unsupervised model for sequence anomaly detection. • Propose a novel spectral regularizer to restrict latent spaces of autoencoders. • Statistical analysis on DWT coefficients to mine unique features of normal sequences. • Design a Weight Controller to gain sample-adaptive regularization weights. Nowadays, systems or entities are usually monitored by devices, generating large amounts of time series. Detecting anomalies in them help prevent potential losses, thus arousing much research interest. Existing studies have adopted autoencoders to detect anomalies, where reconstruction errors are used to indicate outliers. However, sometimes autoencoders may also reconstruct anomalies well due to the learned general features in latent spaces. To solve the above problem, we propose to regularize autoencoders to grasp specific features of normal sequences. Specifically, spectral unique patterns are captured by statistical analysis on discrete wavelet transform (DWT) coefficients of input sequences, restricting latent spaces to reflect unique patterns of normal sequences in both time and frequency domains. Furthermore, a Weight Controller calculating sample-adaptive regularization weights is designed to fully utilize the regularization effect. Extensive experiments on three public benchmarks demonstrate the effectiveness and superiority of the proposed model compared with state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

With continuous queries that are used widely in location based mobile social networking services, how to protect the location privacy effectively for continuous query has been a hot topic for researchers. In this paper, we analyze the existing location privacy protection systems and algorithms for location based services; considering their disadvantages of slow responding time and high anonymization costs, we propose a new enhanced greedy cloaking algorithm which predicts a cloaking area at the initial query to be the cloaking region in the whole query lifetime by the comprehensive computation of privacy monitor, quality monitor, and dynamic adjuster. Privacy monitor and quality monitor charge the privacy protection level and service quality degree respectively; dynamic adjuster can adjust the cycle center point dynamically. We employ cycle as cloaking region form which can effectively alleviate the computation overhead. And we compare it with the earlier algorithm on three aspects. The experimental result shows that the enhanced greedy cloaking algorithm is better than the original greedy algorithm on average responding time or anonymization cost. [ABSTRACT FROM AUTHOR]

Traffic flow prediction (TFP) has attracted increasing attention with the development of smart city. In the past few years, neural network-based methods have shown impressive performance for TFP. However, most of previous studies fail to explicitly and effectively model the relationship between inflows and outflows. Consequently, these methods are usually uninterpretable and inaccurate. In this paper, we propose an interpretable local flow attention (LFA) mechanism for TFP, which yields three advantages. (1) LFA is flow-aware. Different from existing works, which blend inflows and outflows in the channel dimension, we explicitly exploit the correlations between flows with a novel attention mechanism. (2) LFA is interpretable. It is formulated by the truisms of traffic flow, and the learned attention weights can well explain the flow correlations. (3) LFA is efficient. Instead of using global spatial attention as in previous studies, LFA leverages the local mode. The attention query is only performed on the local related regions. This not only reduces computational cost but also avoids false attention. Based on LFA, we further develop a novel spatiotemporal cell, named LFA-ConvLSTM (LFA-based convolutional long short-term memory), to capture the complex dynamics in traffic data. Specifically, LFA-ConvLSTM consists of three parts. (1) A ConvLSTM module is utilized to learn flow-specific features. (2) An LFA module accounts for modeling the correlations between flows. (3) A feature aggregation module fuses the above two to obtain a comprehensive feature. Extensive experiments on two real-world datasets show that our method achieves a better prediction performance. We improve the RMSE metric by 3.2%–4.6%, and the MAPE metric by 6.2%–6.7%. Our LFA-ConvLSTM is also almost 32% faster than global self-attention ConvLSTM in terms of prediction time. Furthermore, we also present some visual results to analyze the learned flow correlations. [ABSTRACT FROM AUTHOR]

Discovering global knowledge from distributed data sources is challenging as there exist several practical concerns such as bandwidth limitation and data privacy. By appropriately abstracting distributed data, various global data mining tasks could still be implemented on the basis of local data abstractions. This article reviews existing techniques related to distributed data mining in abstraction-based data mining. It then discusses open research challenges on mining tasks performed on distributed and abstracted data, describes how global data models (clustering and manifold discovery) could be learnt based on local data models, and points out future research directions. [ABSTRACT FROM AUTHOR]

Natural disasters caused by heavy rainfall often cause huge loss of life and property. Hence, the task of precipitation nowcasting is of great importance. To solve this problem, several deep learning methods have been proposed to forecast future radar echo images, and then the predicted maps are converted to the distribution of rainfall. The prevailing spatiotemporal sequence prediction methods apply a ConvRNN structure, which combines the convolution and recurrent neural network. Although ConvRNN methods achieve remarkable success, they do not capture both local and global spatial features simultaneously, which degrades the nowcasting in regions of heavy rainfall. To address this issue, we propose a Region Attention Block (RAB) and embed it into ConvRNN to enhance forecasting in the areas with heavy rainfall. Besides, the ConvRNN models find it hard to memorize longer historical representations with limited parameters. To this end, we propose a Recall Attention Mechanism (RAM) to improve the prediction. By preserving longer temporal information, RAM contributes to the forecasting, especially in the moderate rainfall intensity. The experiments show that the proposed model, Region Attention Predictive Network (RAP-Net), significantly outperforms state-of-the-art methods. [ABSTRACT FROM AUTHOR]

The near-surface temperature prediction (NTP) is an important spatial–temporal forecast problem, which can be used to prevent temperature crises. Most of the previous approaches fail to explicitly model the long- and short-range spatial correlations simultaneously, which is critical to making an accurate temperature prediction. In this study, both long- and short-range spatial correlations are captured to fill this gap by a novel convolution operator named Long- and Short-range Convolution (LS-Conv). The proposed LS-Conv operator includes three key components, namely, Node-based Spatial Attention (NSA), Long-range Adaptive Graph Constructor (LAGC), and Long- and Short-range Integrator (LSI). To capture long-range spatial correlations, NSA and LAGC are proposed to evaluate node importance aiming at auto-constructing long-range spatial correlations, which is named as Long-range aware Graph Convolution Network (LR-GCN). After that, the Short-range aware Convolution Neural Network (SR-CNN) accounts for the short-range spatial correlations. Finally, LSI is proposed to capture both long- and short-range spatial correlations by intra-unifying LR-GCN and SR-CNN. Upon the proposed LS-Conv operator, a new model called Long- and Short-range for NPT (LS-NTP) is developed. Extensive experiments are conducted on two real-world datasets and the results demonstrate that the proposed method outperforms state-of-the-art techniques. The source code is available on GitHub: https://github.com/xuguangning1218/LS_NTP. • We jointly incorporate long- and short-range spatial correlations in NTP prediction. • Theoretical findings proved that the proposed LS-Conv is a general version of CNN. • We develop a new spatial–temporal model named LS-NTP for the NTP task. • Extensive experiments are conducted on two real-world datasets. [ABSTRACT FROM AUTHOR]

Precipitation nowcasting aims to predict the rainfall distribution within a short-term period. However, it pays the same attention to all locations instead of emphasizing those regions with heavy rainfall that has more threats to human activity. Therefore, we develop an important task named Heavy Rainfall Forecast (HRF), which mainly focuses on the movement and change of heavy rainfall areas. It sets aside one hour to give meteorological administration sufficient time to issue warning information. To tackle this task, firstly, we rebuild the meteorological radar dataset based on three criteria to obtain the samples involving heavy rainfall. Secondly, we propose the Location-Refining (LR) neural network to combine the advantages of the optical flow-based and deep learning-based methods in predicting higher intensity and more accurate position, respectively. LR neural network consists of a location network and a refining network. The former is responsible for the accurate predictions of position and trend of rainfall, and the later accounts for more accurately estimating the intensity. To make the model pay more attention to the high echo region, we design new loss functions and introduce auxiliary information of high echo values. A series of experiments show that our model has a significant improvement on this task. Specifically, compared with existing methods, we improve the valid mean square error by 6.4% for the threshold being 20 and 15.1% for the threshold being 30. The critical success indexes are improved by 12.8% for the threshold being 20 and 24.8% for the threshold being 30. We also improve the heidke skill score by 9.9% for the threshold being 20 and 21.4% for the threshold being 30. Furthermore, the proposed framework can be well transferred to other deep learning-based models, and improves their performance. • An important task for predicting heavy rainfall and criteria for building datasets. • A novel framework that effectively combines the advantages of two kinds of methods. • Focus more on heavy rainfall with a more reasonable optimization strategy. • Excellently transferred to other models and improve their performance. • Better practical performance for forecasting rainfall, especially for heavy rainfall. [ABSTRACT FROM AUTHOR]

Precipitation nowcasting is an indispensable task for traffic routing and disaster avoidance. Due to its strenuous movement, even the most recent deep learning techniques in computer vision deliver unsatisfactory performance. The main reason can be attributed to the following two aspects: 1) The traditional convolution in the input has a limited field to extract spatial representation. 2) The convolution in state-to-state connection might lead to mismatch problems and inaccurate strength predictions. To address the two drawbacks, we propose an innovative algorithm the Reconstitute Spatiotemporal LSTM (RST-LSTM) based on Convolutional Recurrent Neural Network (ConvRNN). In the proposed model, we present the local and global reconstitution scheme into the current input and the hidden state respectively. The local reconstitution (LR) can adaptively adjust the perceptual field of convolution so as to extract more useful spatial information and exclude invalid representation. Moreover, in the hidden state, the global reconstitution (GR) is embedded to alleviate the problem of mismatching between the current input and hidden state. Experimental results in MovingMNIST++ show that our approach can achieve the best predictions for those data with more drastic changes in the adjacent time. For radar data in CIKM AnalytiCup 2017 (we name it RadarCIKM in this paper), our method outperforms the state-of-the-art competitors. Furthermore, we notice that GR can promote the nowcasting in the high radar echo region and LR can reduce the error. [ABSTRACT FROM AUTHOR]

Few-shot remote sensing scene classification (FSRSSC) is an important task, which aims to recognize novel scene classes with few examples. Recently, several studies attempt to address the FSRSSC problem by following few-shot natural image classification methods. These existing methods have made promising progress and achieved superior performance. However, they all overlook two unique characteristics of remote sensing images: 1) object cooccurrence that multiple objects tend to appear together in a scene image and 2) object spatial correlation that these cooccurrence objects are distributed in the scene image following some spatial structure patterns. Such unique characteristics are very beneficial for FSRSSC, which can effectively alleviate the scarcity issue of labeled remote sensing images since they can provide more refined descriptions for each scene class. To fully exploit these characteristics, we propose a novel scene-graph-matching-based meta-learning framework for FSRSSC, called SGMNet. In this framework, a scene graph construction module is carefully designed to represent each test remote sensing image or each scene class as a scene graph, where the nodes reflect these cooccurrence objects; meanwhile, the edges capture the spatial correlations between these cooccurrence objects. Then, a scene graph matching module is further developed to evaluate the similarity score between each test remote sensing image and each scene class. Finally, based on the similarity scores, we perform the scene class prediction via a nearest neighbor classifier. We conduct extensive experiments on the UCMercedLandUse, WHU19, AID, and NWPU-RESISC45 datasets. The experimental results show that our method obtains superior performance over the previous state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Regional wind speed prediction is an important spatiotemporal prediction problem which is crucial for optimizing wind power utilization. Nevertheless, the complex dynamics of wind speed pose a formidable challenge to prediction tasks. The evolving dynamics of wind could be governed by underlying physical principles that can be described by partial differential equations (PDE). This study proposes a novel approach called PDE-assisted network (PaNet) for regional wind speed prediction. In PaNet, a new architecture is devised, incorporating both PDE-based dynamics (PDE dynamics) and unknown dynamics. Specifically, this architecture establishes interactions between the two dynamics, regulated by an inter-dynamics communication unit that controls interactions through attention gates. Additionally, recognizing the significance of the initial state for PDE dynamics, an adaptive frequency-gated unit is introduced to generate a suitable initial state for the PDE dynamics by selecting essential frequency components. To evaluate the predictive performance of PaNet, this study conducts comprehensive experiments on two real-world wind speed datasets. The experimental results indicated that the proposed method is superior to other baseline methods. [ABSTRACT FROM AUTHOR]

Cloud detection is the task of detecting cloud areas in remote sensing images (RSIs), and it has attracted extensive research interest. Recently, deep learning (DL)-based methods have been proposed and achieved great performance for cloud detection. However, due to the satellite’s limitation in storage and memory, existing DL approaches, which suffer from extensive computation and large model size, are almost impossible to be deployed on satellites. To fill this gap, we target at studying effective and efficient cloud detection solutions that are suitable for satellites. In this article, we develop a lightweight autoencoder-based cloud detection method, namely, lightweight encoder–decoder cloud detection network (LWCDnet). In the encoder part, the designed novel lightweight dual-branch block (LWDBB) in the backbone extracts spatial and contextual information concurrently. Moreover, a lightweight feature pyramid module (LWFPM) is proposed to capture high-level multiscale contextual information. In the decoder part, the lightweight feature fusion module (LWFFM) compensates for the missing spatial and detail information from the encoder to the high-level feature maps. We evaluate the proposed method on two public datasets: LandSat8 and Moderate-Resolution Imaging Spectroradiometer (MODIS). Extensive experiments demonstrate that the proposed LWCDnet achieves comparable accuracy as the state-of-the-art cloud detection methods and lightweight semantic segmentation algorithms. In the meantime, LWCDnet has much less computation burden with smaller model size. [ABSTRACT FROM AUTHOR]

Twitter is widely used all over the world, and a huge number of hot topics are generated by Twitter users in real time. These topics are able to reflect almost every aspect of people’s daily lives. Therefore, the detection of topics in Twitter can be used in many real applications, such as monitoring public opinion, hot product recommendation and incidence detection. However, the performance of traditional topic detection methods is still far from perfect largely owing to the tweets’ features, such as their limited length and arbitrary abbreviations. To address these problems, we propose a novel framework (MVTD) for Twitter topic detection using multiview clustering, which can integrate multirelations among tweets, such as semantic relations, social tag relations and temporal relations. We also propose some methods for measuring relations among tweets. In particular, to better measure the semantic similarity of tweets, we propose a new document similarity measure based on a suffix tree (STVSM). In addition, a new keyword extraction method based on a suffix tree is proposed. Experiments on real datasets show that the performance of MVTD is much better than that of a single view, and it is useful for detecting topics from Twitter. [ABSTRACT FROM PUBLISHER]

Typhoons are one of the most destructive types of disasters. Several statistical models have been designed to predict their paths to reduce damage, casualties, and economic loss. To further increase prediction accuracy, two key challenges are (1) to extract better nonlinear 3D features of typhoons, which is hard due to their complex high-dimensional properties, and (2) to combine suitable 2D and 3D features in a proper way to improve predictions. To address these challenges, this paper presents a novel spatio-temporal deep learning model named Attention-based Multi ConvGRU (AM-ConvGRU). To automatically select high response isobaric planes of typhoons when considering their whole 3D structures, AM-ConvGRU leverages the Residual Channel Attention Block (RCAB). Furthermore, it integrates a novel model named Multi-ConvGRU to extract large-scale nonlinear spatial features of typhoons. Moreover, the approach relies on a Wide & Deep framework to fuse the traditional Generalized Linear Model (GLM) with the proposed AM-ConvGRU model. To evaluate the designed approach, extensive experiments have been conducted using real-world typhoons data from the Western North Pacific (WNP) basin obtained from both the China Meteorological Administration (CMA) dataset and the EAR-Interim dataset maintained by the European Centre for Medium-Range Weather Forecasts (ECMWF). Results show that the proposed method outperforms state-of-the-art deep learning typhoon prediction methods. The source code is available on GitHub with the following link:https://github.com/xuguangning1218/Typhoon_Path. [ABSTRACT FROM AUTHOR]

• The SAF-Net is proposed to combine domain-knowledge and data-driven knowledge. • The spatial attention is used to automatically selected high-response wind speed area. • Extensive experiments is proceeded in the real-world dataset. A typhoon is a destructive weather system that can cause severe casualties and economic losses. Typhoon intensity (TI) is a measurement to evaluate its ruinous degree. Hence, typhoon intensity prediction is an important research problem and many methods have been proposed. However, most of the existing approaches have very limited capability to combine the 2D Typhoon Structure Domain-expert Knowledge (2D-TSDK) and the 3D Typhoon Structure Data-driven Knowledge (3D-TSDK) for the TI prediction. To address this issue, this paper proposes a spatio-temporal deep learning method named Spatial Attention Fusing Network (SAF-Net). The designed model aims to fuse the 2D-TSDK and the 3D-TSDK by developing a specific Wide & Deep framework. In the data-driven component, a special Spatial Attention (SA) module is designed to automatically select high-response wind speed areas and embedded into a three-branch CNN to exploit the 3D-TSDK. Then, the Wide & Deep framework integrates the 2D-TSDK and the 3D-TSDK for the TI prediction. Comprehensive experiments have been conducted on a real-world dataset, and the result shows that the proposed method outperforms state-of-the-art typhoon intensity prediction methods. The code is available in GitHub: https://github.com/xuguangning1218/TI_Prediction [ABSTRACT FROM AUTHOR]

Abstract: In this paper, we study the problem of multi-label collective classification (MLCC) where instances are related and associated with multiple class labels. Such correlation of class labels among interrelated instances exists in a wide variety of data, e.g., a web page can belong to multiple categories since its semantics can be recognized in different ways, and the linked web pages are more likely to have the same classes than the unlinked pages. We propose an effective and novel Markov chain based learning method for MLCC problems. Our idea is to model the problem as a Markov chain with restart on transition probability graphs, and to propagate the ranking score of labeled instances to unlabeled instances based on the affinity among instances. The affinity among instances is set up by explicitly using the attribute features derived from the content of instances as well as the correlation features constructed from the links of instances. Intuitively, an instance which contains linked neighbors that are highly similar to the other instances with a high rank of a particular class label, has a high chance of this class label. Extensive experiments have been conducted on two DBLP datasets to demonstrate the effectiveness of the proposed algorithm. The performance of the proposed algorithm is shown to be better than those of the binary relevance multi-label algorithm, collective classification algorithms (wvRN, ICA and Gibbs), and the ICML algorithm for the tested MLCC problems. [Copyright &y& Elsevier]

The main aim of this paper is to develop a community discovery scheme in a multi-dimensional network for data mining applications. In online social media, networked data consists of multiple dimensions/entities such as users, tags, photos, comments, and stories. We are interested in finding a group of users who interact significantly on these media entities. In a co-citation network, we are interested in finding a group of authors who relate to other authors significantly on publication information in titles, abstracts, and keywords as multiple dimensions/entities in the network. The main contribution of this paper is to propose a framework (MultiComm)to identify a seed-based community in a multi-dimensional network by evaluating the affinity between two items in the same type of entity (same dimension)or different types of entities (different dimensions)from the network. Our idea is to calculate the probabilities of visiting each item in each dimension, and compare their values to generate communities from a set of seed items. In order to evaluate a high quality of generated communities by the proposed algorithm, we develop and study a local modularity measure of a community in a multi-dimensional network. Experiments based on synthetic and real-world data sets suggest that the proposed framework is able to find a community effectively. Experimental results have also shown that the performance of the proposed algorithm is better in accuracy than the other testing algorithms in finding communities in multi-dimensional networks. [ABSTRACT FROM AUTHOR]

Sequence anomaly detection in time series is of critical importance to wide applications ranging from finance, healthcare to IT system monitoring. Most current researches use the reconstruction-based deep learning algorithms to solve the problem. In this article, we aim to use a prediction-based method to detect sequence anomalies in univariate time series, because the latter methods can detect anomalies using historical information revealing normal patterns in time series whereas the former methods simply consider current sequences. However, it is challenging because there exists both uncertainty in the future and performance deterioration under long detection horizon. To tackle the challenges, we propose an unsupervised algorithm called KfreqGAN, which is based on adversarially trained sequence predictor. The adversarial learning architecture helps the model make accurate predictions for future sequences. In addition, auxiliary information from frequency domain is used to help the model capture the characteristics of time series for achieving satisfactory predictions. We conduct extensive experiments on two public-available datasets, with results demonstrating the effectiveness of the proposed algorithm and its superiority to baseline algorithms. [ABSTRACT FROM AUTHOR]

Reinforcement learning (RL) techniques have recently been introduced to recommender systems. Most existing research works focus on designing policy and learning algorithms of the recommender agent but seldom care about the top-aware issue , i.e., the performance on the top positions is not satisfying, which is crucial for real applications. To address the drawback, we propose a Supervised deep Reinforcement learning Recommendation framework named as SRR. Within this framework, we utilize a supervised learning (SL) model to partially guide the learning of recommendation policy, where the supervision signal and RL signal are jointly employed and updated in a complementary fashion. We empirically find that suitable weak supervision helps to balance the immediate reward and the long-term reward, which nicely addresses the top-aware issue in RL based recommendation. Moreover, we perform a further investigation on how different supervision signals impact on recommendation policy. Extensive experiments are carried out on two real-world datasets under both the offline and simulated online evaluation settings, and the results demonstrate that the proposed methods indeed resolve the top-aware issue without much performance sacrifice in the long-run, compared with the state-of-the-art methods. [ABSTRACT FROM AUTHOR]

Accurate hyperspectral image classification has been an important yet challenging task for years. With the recent success of deep learning in various tasks, 2-dimensional (2D)/3-dimensional (3D) convolutional neural networks (CNNs) have been exploited to capture spectral or spatial information in hyperspectral images. On the other hand, few approaches make use of both spectral and spatial information simultaneously, which is critical to accurate hyperspectral image classification. This paper presents a novel Synergistic Convolutional Neural Network (SyCNN) for accurate hyperspectral image classification. The SyCNN consists of a hybrid module that combines 2D and 3D CNNs in feature learning and a data interaction module that fuses spectral and spatial hyperspectral information. Additionally, it introduces a 3D attention mechanism before the fully-connected layer which helps filter out interfering features and information effectively. Extensive experiments over three public benchmarking datasets show that our proposed SyCNNs clearly outperform state-of-the-art techniques that use 2D/3D CNNs. [ABSTRACT FROM AUTHOR]

Conventionally, it is a prerequisite to acquire a good number of annotated data to train an accurate classifier. However, the acquisition of such dataset is usually infeasible due to the high annotation cost. Therefore, semi-supervised learning has emerged and attracts increasing research efforts in recent years. Essentially, semi-supervised learning is sensitive to the manner how the unlabeled data is sampled. However, the model performance might be seriously deteriorated if biased unlabeled data is sampled at the early stage. In this paper, an unbiased semi-supervised cluster tree is proposed which is learnt using only very few labeled data. Specifically, a K-means algorithm is adopted to build each level of this hierarchical tree in a decent top-down manner. The number of clusters is determined by the number of classes contained in the labeled data. The confidence error of the cluster tree is theoretically analyzed which is then used to prune the tree. Empirical studies on several datasets have demonstrated that the proposed semi-supervised cluster tree is superior to the state-of-the-art semi-supervised learning algorithms with respect to classification accuracy. [ABSTRACT FROM AUTHOR]