Search

The Black Hole Clustering (BHC) algorithm is a density-based partitional clustering method inspired by the Density-based Spatial Clustering of Applications with Noise (DBSCAN). It does not require the number of clusters nor the computation of the pair-wise distance matrix between the data points, making it faster than DBSCAN. Also, it only needs one parameter that is intuitively easier to set than the epsilon parameter of DBSCAN. However, BHC needs the allocation of the so-called black holes that have to be linearly independent, making the algorithm in its current version suitable only for two or three-dimensional data sets. In this paper, we propose a generalized version of the black hole clustering algorithm (GBHC) by introducing a novel black hole allocation procedure for higher-dimensional data spaces. Furthermore, the proposed method is data-independent, so we have to run it once to obtain the black hole positions for all finite-dimensional metric spaces. We performed extensive computational experiments to compare GBHC with DBSCAN. The results show that both algorithms obtain comparable clustering solutions. GBHC, however, outperforms DBSCAN in computational complexity and explainability. • We propose a novel method to place the black holes for high-dimensional data sets. • This method is called Generalized Black Hole Clustering and it is data-independent. • We run the new method once to get black hole positions in all finite metric spaces. • We compare the GBHC and DBSCAN algorithms using several validation measures. [ABSTRACT FROM AUTHOR]

• Pipeline to extract information from historical tables with different layouts. • Two information extraction systems are compared at each phase of the pipeline. • Deal with multispan and classify headers by content. • Improvements in both systems achieve new state-of-the-art results in HisClima tables. Document Image Understanding is a demanding Pattern Recognition problem that requires complex recognition models. This problem is even more difficult for document images with complicated layouts like tables, where the reading order is often intrinsically ambiguous, and consequently, the context is generally ambiguous as well. In this paper, we compare two machine learning approaches for extracting information in pre-printed historical tables with handwritten information. We analyze the performance of each approach at each step of the extraction process over different corpora, up to a realistic scenario where documents with different table layouts written by different hands are used. The results are good in general and show that a model based on Multilayer Perceptrons yields better results on more homogeneous documents, while another model based on Graph Neural Networks generalizes better on heterogeneous corpora. [ABSTRACT FROM AUTHOR]

• A word embedding-based topic model (WETM) for short text documents. • Sparsity problem removed in short text and discovered structural information for topics and words. • A modified collapsed Gibbs sampling algorithm to find the parameters for WETM. • WETM achieved better classification, topic coherence, topic quality, and clustering results. • The execution time is lower for WETM as compared to baseline topic models. Short texts are a common source of knowledge, and the extraction of such valuable information is beneficial for several purposes. Traditional topic models are incapable of analyzing the internal structural information of topics. They are mostly based on the co-occurrence of words at the document level and are often unable to extract semantically relevant topics from short text datasets due to their limited length. Although some traditional topic models are sensitive to word order due to the strong sparsity of data, they do not perform well on short texts. In this paper, we propose a novel word embedding-based topic model (WETM) for short text documents to discover the structural information of topics and words and eliminate the sparsity problem. Moreover, a modified collapsed Gibbs sampling algorithm is proposed to strengthen the semantic coherence of topics in short texts. WETM extracts semantically coherent topics from short texts and finds relationships between words. Extensive experimental results on two real-world datasets show that WETM achieves better topic quality, topic coherence, classification, and clustering results. WETM also requires less execution time compared to traditional topic models. [ABSTRACT FROM AUTHOR]

• Large-scale experimental comparative analysis of the newer and most influential multi-class classification algorithms. • Large-scale experimental comparative analysis of the binary classifier combination method. • Collaborative prediction of target location using integrated learning methods. In recent years, high spatial resolution remote sensing technology has made significant progress. High-resolution remote sensing satellites provide great convenience for high-quality image acquisition. In order to adapt to changes in the appearance of the target, mainstream tracking algorithms often use pattern recognition methods to build a target appearance model with learning capabilities, and use the image frames acquired during the tracking process to update the appearance model. This paper mainly studies the object-oriented remote sensing image information extraction method based on multi-classifier combination and deep learning algorithm. In this paper, we use the splitting mechanism of the tree structure to retain the appearance model with diversity, and through the integrated learning integration strategy, the target position is collaboratively predicted. Through the comparative analysis on the OTB and VOT platforms, the algorithm works well when the requirements of the tracking standards are low (the accuracy threshold is greater than 20 pixels and the success threshold is less than 0.4 pixels). The experimental results in this paper show that compared with other advanced classification methods, the proposed method shows better generalization performance in accuracy, recall, f-measure, g-mean and AUC. [ABSTRACT FROM AUTHOR]

• The paper proposes a new feature selection model and supervised learning method. • The method can select important samples. • The method greatly improve the efficiency and stability of the algorithm. Feature selection is an effective method to reduce the feature dimension and improve the efficiency of data mining algorithms, it can be used to solve the "curse of dimensionality" problem. Most existing feature selection methods do not adequately consider the effects of noise and outliers, thereby reducing the performance of regression. The paper proposed a new feature selection model and supervised learning method, using self-paced regularizer (SP-regularizer) terms and ℓ 2, 1 -norm to constrain the model. Specifically, the feature selection model is trained firstly using the most high-confidence samples base on the self-paced learning theory, and then adds the more high-confidence training samples in the remaining samples to increase the generalization ability of the initial feature selection model until the generalization ability of the feature selection model is not improved or all training samples are used up. Then use ℓ 2, 1 -norm remove redundant feature to improve algorithm efficiency. Experimental results on nine datasets show that our proposed method is superior to the comparison algorithms. [ABSTRACT FROM AUTHOR]

• A general model is proposed for improving the performance of traditional methods. • Shear transformation enhances the directional features of traditional methods. • The general model has better performances to extract directional linear information. • The general model has very strong universality. Most images have lots of linear information, which plays an important role in the tasks of image processing and pattern recognition. However, due to the interference of complex background in the images, the multiple directional characteristics of the linear information, and the problems of directional limitations in traditional methods, all these lead to the incomplete and discontinuities existed in the extracted linear information. To solve these problems, this paper puts forward a general model for improving the performance of the linear information extraction methods (LIEM) by utilizing the shear transformation. In this model, the shear transformation can transform an object (a filter or an image) in multiple directions, which directly or indirectly increases the directional characteristics of the traditional LIEM, and improves their performances to extract directional linear information and weak linear information. This paper elaborates on the basic principles of the model and its two implementations for the specific tasks. Furthermore, a variety of experiments are made to verify the versatility and effectiveness of the proposed model. [ABSTRACT FROM AUTHOR]

• Anomaly detection is one of the most common problems in data science. • Unsupervised approaches are useful since anomalies are not easily reproducible. • Triangle-based outlier detection is based on geometric reasoning. • Results achieved show a higher robustness than state of the art alternatives. For the last decades, anomaly detection has been one of the most common problems in data mining and computer science projects. The scientific community has made a great effort to develop methods and techniques for the detection of elements that deviate from the norm. Many of these techniques follow an unsupervised approach since anomalies are scarce and not easily reproducible. In this paper, a novel unsupervised anomaly detection method based on geometric reasoning is proposed. Triangle-based Outlier Detection (TOD) falls within the group of distance-based anomaly detection techniques and has been compared to state of the art methods for unsupervised anomaly detection. Results achieved prove that TOD offers greater robustness than alternative methods for the selection of the decision thresholds that condition the detection of outliers. [ABSTRACT FROM AUTHOR]

• Clustering log data is fundamental for providing insights into many critical areas. • Feedback such as labels and pairwise constraints can improve log clustering quality. • A principled way to summarize the data can be used to deal with a high volume of data. Machine-generated log data can provide valuable insights into many critical areas such as system failures, network security, and performance optimization. The increasing prominence of this data in both volume and complexity requires data mining approaches that are both scalable and flexible. In this paper, we propose a new approach for clustering machine-generated logs which contains a novel combination of the use of the coreset with user feedback. The coreset allows us to efficiently summarize the data in a principled manner such that performance after fitting model parameters on the coreset is similar to the performance that would have been achieved with the full dataset. Furthermore, the formal approach we propose allows users to incorporate two different types of feedback, in the forms of labels and pairwise constraints, to further improve results and better deal with the increasing complexity and variety of log datasets. [ABSTRACT FROM AUTHOR]

• End-to-end Transformer-based architecture for named entity recognition in handwritten document images. • Joint recognition of handwriting and named entity at the paragraph level, allowing the model to avoid segmentation errors. • Exploration of different training scenarios including two-stage, mixed-data, and curriculum learning. • Extensive ablation and comparative experiments are conducted to validate the effectiveness of our approach. The extraction of relevant information carried out by named entities in handwriting documents is still a challenging task. Unlike traditional information extraction approaches that usually face text transcription and named entity recognition as separate subsequent tasks, we propose in this paper an end-to-end transformer-based approach to jointly perform these two tasks. The proposed approach operates at the paragraph level, which brings two main benefits. First, it allows the model to avoid unrecoverable early errors due to line segmentation. Second, it allows the model to exploit larger bi-dimensional context information to identify the semantic categories, reaching a higher final prediction accuracy. We also explore different training scenarios to show their effect on the performance and we demonstrate that a two-stage learning strategy can make the model reach a higher final prediction accuracy. As far as we know, this work presents the first approach that adopts the transformer networks for named entity recognition in handwritten documents. We achieve the new state-of-the-art performance in the ICDAR 2017 Information Extraction competition using the Esposalles database, for the complete task, even though the proposed technique does not use any dictionaries, language modeling, or post-processing. [ABSTRACT FROM AUTHOR]

• Three-way concept stability and its properties. • Algorithm for the computation of three-way concept stability. • A case study on natural language generation scenario. Three-way concept analysis (3WCA) has been an emerging and important methodology for knowledge discovery and data analysis. Particularly, 3WCA can efficiently characterize the information of "jointly possessed" and "jointly not possessed" compared to the classical formal concept only can describe common attributes owned by objects. This property, typical of 3WCA has a huge potential in the field of Natural Language Generation (NLG). However, the construction of a three-way concept lattice is proved as an NP-complete problem and even harder than the construction of conventional concept lattice. This could negatively affect the use of 3WCA for NLG in real contexts. Hence, it is necessary to prune the three-way concept lattice and extract more interesting three-way concepts for knowledge acquisition. To this end, this paper defines the stability of a three-way concept and analyzes the relevant properties. An efficient computational algorithm for calculating the stability of three-way concepts is developed and evaluated by an experiment. In addition, a case study on NLG is conducted for demonstrating the applicability of the proposed technique. [ABSTRACT FROM AUTHOR]

• Reading grid of six properties for regression filter criteria analysis. • Analysis of a nonlinear coefficient of determination criterion for feature selection. • Comparison of three relevance criteria for regression problems. Feature selection is an important preprocessing step in machine learning. It helps to better understand the importance of some features and to reduce the dimensionality of a dataset, which improves machine learning and information extraction. Among the different existing methods for selecting features, filters are popular because they are independent from the model, which will be learnt afterwards, and computationally efficient. The efficiency of filter methods relies on a strategic choice: the choice of the relevance criterion. Many criteria exist; they exhibit various properties, which in turn result in selecting different features. The choice of the criterion is thus important and should ideally be linked to the properties of the data and to users' goals. This paper shows that six properties should be analysed when selecting a relevance criterion in the context of regression problems. It proposes a reading grid to analyse relevance criteria and to make a well-guided choice. [ABSTRACT FROM AUTHOR]

As an important tool of data mining, clustering analysis can measure similarity between different data and classify them. It is widely applied in many fields such as pattern recognition, economics and biology. In this paper, we propose a new clustering algorithm. First, original unclassified dataset is converted into a weighted complete graph in which a node represents a data point and distance between two data points is used as weight of the edge between the corresponding two nodes. Second, local importance of each node in the network is calculated and evaluated by Laplacian centrality. The cluster center has higher Laplacian centrality than surrounding neighbor nodes and relatively large distance from nodes with higher Laplacian centralities. The new algorithm is a true parameter-free clustering method. It can automatically classify the dataset without any priori parameters. In this paper, the new algorithm was compared with 8 well-known clustering algorithms in 7 real datasets. Results show that the proposed algorithm has good clustering effect. [ABSTRACT FROM AUTHOR]

• A novel Particle swarm optimization based algorithm (PSO-GES), is introduced. • A guided exploration strategy in order to generate high-quality rules, is proposed. • A new fitness funtion for obtain the fitness value in a shorter runtime, is introduced. • PSO-GES uses a summary matrix with a smaller number of rows than the original database. • PSO-GES mines a set of better quality rules than the association rules mining algorithms compared. Association rule mining is one of the most important and active research areas in data mining. In the literature, several association rule miners have been proposed; among them, those based on particle swarm optimization (PSO) have reported the best results. However, these algorithms tend to prematurely fall into local solutions, avoiding a wide exploration that could produce even better results. In this paper, an algorithm based on PSO, called PSO-GES, for mining association rules using a Guided Exploration Strategy is introduced. Our experiments, over real-world transactional databases, show that our proposed algorithm mines better quality association rules than the most recent PSO-based algorithms for mining association rules of the state of the art. [ABSTRACT FROM AUTHOR]

• Multiscale Egocentric Summarization and Action Ranking with hierarchical clustering. • Proposed solution follows an "Analyze once, Generated many" principle. • Actions ranked from occurrences of action frames in multiscale summary hierarchy. Useful information extraction from egocentric videos has evolved as an important research problem for both computer vision and multimedia communities. In this paper, we have addressed two problems, namely (i) generating multiscale summaries, i.e., multiple summaries of different lengths and (ii) priority-based ranking of various actions present in egocentric videos. A new algorithm, termed as Multiscale Egocentric Video Summarization and Action Ranking (MEVSAR), with agglomerative clustering as its backbone, is proposed to solve the above problems. Importantly, the MEVSAR algorithm follows an "analyze once, generate many" principle to generate multiple summaries in a single run and subsequently rank actions from the generated summaries. Experimental evaluation on two well-known publicly available datasets clearly demonstrate the merits of the proposed approach. [ABSTRACT FROM AUTHOR]

• An algorithm for multi-core frequent itemsets mining in data streams is presented. • It is proposed a processing scheme which allows to process data streams transmitted at different rates. • The Gearman framework is used to implement the proposed multi-core processing algorithm. • Hashing and lexicographic order of received items are used for frequent itemsets mining in data streams. Data streams are modern data sources that are gaining attention as a consequence of their many practical applications (they can be found in data transmission, eCommerce, and intrusion detection system among others). Nevertheless, the efforts to obtain insights from data streams are limited due to their massive information volume and the time needed to process them. In this paper, a new approach for Frequent Itemsets Mining on data streams based on prefix trees which takes advantage of multi-core systems is proposed. This approach uses the Gearman framework as the interface for multi-core processing, and it allows to exploit their scalability efficiently. Experimental results show that the proposed method obtains the same patterns compared with similar approaches reported in the state-of-the-art and outperforms them concerning the processing time required. Also, it is proved that the proposed method is insensitive to variations in the support threshold value, and its efficiency depends on the size of the transactions and not on the size of the alphabet, which is a significant issue in other Frequent Itemsets Mining algorithms. [ABSTRACT FROM AUTHOR]

• Large Scale Image to Video Retrival. • Convolutional Neural Networks. • Bag of Visual Word. • Visual Weighted Inverted Index. This paper aims to solve the problem of large-scale video retrieval by a query image. Firstly, we define the problem of top- k image to video query. Then, we combine the merits of convolutional neural networks(CNN for short) and Bag of Visual Word(BoVW for short) module to design a model for video frames information extraction and representation. In order to meet the requirements of large-scale video retrieval, we propose a visual weighted inverted index(VWII for short) and related algorithm to improve the efficiency and accuracy of retrieval process. Comprehensive experiments show that our proposed technique achieves substantial improvements (up to an order of magnitude speed up) over the state-of-the-art techniques with similar accuracy. [ABSTRACT FROM AUTHOR]

Highlights • Pyramid network generates multi-scale proposals fast with no more than twice image resizing operations. • Multi-task learning improves the accuracy of face location. • Online and offline hard samples mining strengthen the power of networks. • Increasing convolutional stride saves much computation time in convolutional layers. • Knowledge distilling and multi-layer merging both improve the performance of small models. Abstract Numerous recent face detectors based on convolutional neural networks (CNNs) have significantly improved the detection performance. However, CNNs usually have a huge number of parameters which lead to very low detection speeds. To address this issue, this paper proposes a fast CNNs cascade face detector with multi-task learning and network acceleration techniques. In particular, the first stage of the detector is an elaborately designed fully convolutional network with a novel pyramid architecture, which can generate multi-scale face proposals efficiently with no more than twice image resizing operations. Several network compression and acceleration techniques including multi-layer merging and knowledge distilling are adopted to further improve inference speed. In addition, online and offline hard sample mining are jointly utilized to further strengthen the power of networks. The experimental results on challenging FDDB show that the proposed face detector is comparable in performance with the-state-of-arts while the speed reaches astonishing 165 fps on Titan GPU. [ABSTRACT FROM AUTHOR]

Highlights • Propose a new named-entity pattern recognition method in a syllable bigram units. • Jointly train positional information in the syllable bigram vector representation. • Compared with the existing methods, both speed and performance are improved. Abstract Named-entity recognition, part of information extraction, is the task of finding the position of a proper names in a sentence and assigning it to the correct category. Existing studies have access to Korean named-entity recognition by a morphological-level method that performs named-entity recognition processes by using the results of morphological analysis as input. While this method has the advantage of using various linguistic clues, it suffers from the error propagation problem of morphological analysis. In this paper, we propose an effective method for Korean syllable-level named-entity recognition to solve the above problem. Firstly, we suggest an approach to use the syllable bi-gram vector representation for Korean syllable-level named-entity recognition. Secondly, influenced by the linguistic characteristics of Korean, we suggest a novel way to make the joint vector representation of syllable bi-gram and Korean eojeol's positional information. In the experiment, we have evaluated our methods on the two Korean named-entity recognition corpora using Bi-directional LSTM-CRFs as a sequence labeler. Experimental results verify that our methods significantly improve the performance of syllable-level named-entity recognition and have similar performance to existing morphological-level named-entity recognition. Besides, additional experiments have shown that our syllable-level named-entity recognition is not only more robust but also faster than traditional morphological-level named-entity recognition by eliminating the morphological analysis process. [ABSTRACT FROM AUTHOR]

This paper addresses a relatively new concept: the buzzword. Buzzwords are fashionable words that continue gaining popularity until a tipping point is reached and then their popularity declines. Our goal in this study is to identify buzzwords through their frequency of occurrence over the years, using two clustering techniques: k -means and the self-organizing map (SOM). We also used the DBLP database to run experiments with data from published papers in an attempt to find terms that could be classified as buzzwords, in accordance with the defined meaning. Clusters generated by both k-means and SOM are very similar, indicating that it is very likely that buzzwords were correctly identified as such. We were able to identify terms such as “android” and “mapreduce”, which were clearly buzzwords for 2012, as well as terms such as “pomdp”, which was not an obvious buzzword. As a contribution, we highlight common characteristics identified for buzzwords and make comparisons between the two methods for finding buzzwords which were analyzed in this paper. [ABSTRACT FROM AUTHOR]

A key issue when mining web information is the labeling problem: data are abundant on the web but is unlabeled. In this paper, we address this problem by proposing (i) a granular model that structures categorical noun phrase instances as well as semantically related noun phrase pairs from a given corpus representing unstructured web pages with a tolerance form of rough sets, (ii) a semi-supervised Tolerant Pattern Learning (TPL) algorithm that labels categorical instances as well as relations. This work is an extension of the TPL algorithm presented in our earlier paper. Our model treats noun phrases, which are described as sets of their co-occurring contextual patterns. We use the ontological information from the Never Ending Language Learner (Nell) system. We compared the performance of our algorithm with Coupled Bayesian Sets (CBS) and Coupled Pattern Learner (CPL) algorithms for categorical and relational extractions, respectively. Experimental results suggest that TPL can achieve comparable performance with CBS and CPL in terms of precision. [ABSTRACT FROM AUTHOR]

Highlights • It proposes a new embedded unsupervised feature selection method. • It integrates the global and local information of the data. • It directly ranks features by employing a simplex-based sparse learning strategy. • It adaptively sets a hyper-parameter based on the number of the selected features. • It provides an effective iterative algorithm to solve the formulated problem. Abstract Unsupervised feature selection is a powerful tool to process high-dimensional data, in which a subset of features are selected out for effective data representation. In this paper, we propose a novel unsupervised feature selection method which discovers and exploits the global information of the data by maximizing distances between samples from different clusters, and preserving the locality of the data by incorporating a Laplacian regularization. Moreover, the proposed method directly ranks the features without any transformation by introducing a simplex-based sparse learning strategy, and enables highly discriminative features to be chosen. Extensive experiments are carried out and the results show effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Zero-shot learning (ZSL) aims to recognize classes whose samples did not appear during training. Existing research focuses on mapping deep visual feature to semantic embedding space explicitly or implicitly. However, ZSL improvements led by discriminative feature transformation is not well studied. In this paper, we propose a ZSL framework that maps semantic embeddings to a discriminative representation space, which are learned in two different ways: Kernelized Linear Discriminant Analysis (KLDA) and Central-loss based Network (CLN). KLDA and CLN can both force samples to be intra-class aggregation and inter-class separation. With the learned discriminative representations, we map class embeddings to representation space using Kernelized Ridge Regression (KRR). Our experiments show that both KLDA+KRR and CLN+KRR surpass state-of-art approaches in both recognition and retrieval task. [ABSTRACT FROM AUTHOR]

Previous Spectral Feature Selection (SFS) methods output promising feature selection results in many real-world applications, which deeply depend on the preservation of the local or global structures of the data via learning a graph matrix. However, current SFS methods 1) learn the graph matrix in the original data which may contain a number of noise to affect the results of feature selection, 2) conduct the learning of a low-dimensional feature space and the graph matrix individually, thus hard achieve the optimal results of feature selection even though both of these two steps achieve their individual optimization, and 3) consider either the local or global structure of data to difficult provide complementary information for feature selection. To address the above issues, this paper proposes a novel supervised feature selection algorithm to simultaneously preserve the local structure (via adaptive structure learning in a low-dimensional feature space of the original data) and the global structure (via a low-rank constraint) of the data. Moreover, we also propose a new optimization method to fast optimize the resulting objective function. We finally verify the proposed method on eight real-word and benchmark datasets, by comparing with the state-of-the-art feature selection methods, and experimental results show that our proposed method achieves competitive results in term of classification performance. [ABSTRACT FROM AUTHOR]

Density-based clustering has been widely used in many fields. A new effective grid-based and density-based spatial clustering algorithm, GRIDEN, is proposed in this paper, which supports parallel computing in addition to multi-density clustering. It constructs grids using hyper-square cells and provides users with parameter k to control the balance between efficiency and accuracy to increase the flexibility of the algorithm. Compared with conventional density-based algorithms, it achieves much higher performance by eliminating distance calculations among points based on the newly proposed concept of ε-neighbor cells. Compared with conventional grid-based algorithms, it uses a set of symmetric ( 2 k + 1 ) D cells to identify dense cells and the density-connected relationships among cells. Therefore, the maximum calculated deviation of ε-neighbor points in the grid-based algorithm can be controlled to an acceptable level through parameter k . In our experiments, the results demonstrate that GRIDEN can achieve a reliable clustering result that is infinite closed with respect to the exact DBSCAN as parameter k grows, and it requires computational time that is only linear to N . [ABSTRACT FROM AUTHOR]

In this paper, we review recent progresses in the area of mining data from multiple data sources. The advancement of information communication technology has generated a large amount of data from different sources, which may be stored in different geological locations. Mining data from multiple data sources to extract useful information is considered to be a very challenging task in the field of data mining, especially in the current big data era. The methods of mining multiple data sources can be divided mainly into four groups: (i) pattern analysis, (ii) multiple data source classification, (iii) multiple data source clustering, and (iv) multiple data source fusion. The main purpose of this review is to systematically explore the ideas behind current multiple data source mining methods and to consolidate recent research results in this field. [ABSTRACT FROM AUTHOR]

Named entity recognition (NER) is an important task for biomedical text mining. Most prior work focused on recognizing regular entities that consist of continuous word sequences and are not overlapped with each other. In this paper, we propose a neural network model called Bi-LSTM-CRF that consists of bidirectional (Bi) long short-term memories (LSTMs) and conditional random fields (CRFs) to identify regular entities and the components of irregular entities. Then the components are combined to build final irregular entities according to manually designed rules. Furthermore, we propose a novel model called NerOne that consists of the Bi-LSTM-CRF network and another Bi-LSTM network. The Bi-LSTM-CRF network performs the same task as the aforementioned model, and the Bi-LSTM network determines whether two components should be combined. Therefore, NerOne automatically combines the components instead of using manually designed rules. We evaluate our models on two datasets for recognizing regular and irregular biomedical entities. Experimental results show that, with less feature engineering, the performances of our models are comparable with those of state-of-the-art systems. We show that the method of automatically combining the components is as effective as the method of manually designing rules. Our work can facilitate the research on biomedical text mining. [ABSTRACT FROM AUTHOR]

Nonnegative matrix factorization (NMF) is a powerful tool for data mining. However, the emergence of ‘big data’ has severely challenged our ability to compute this fundamental decomposition using deterministic algorithms. This paper presents a randomized hierarchical alternating least squares (HALS) algorithm to compute the NMF. By deriving a smaller matrix from the nonnegative input data, a more efficient nonnegative decomposition can be computed. Our algorithm scales to big data applications while attaining a near-optimal factorization, i.e., the algorithm scales with the target rank of the data rather than the ambient dimension of measurement space. The proposed algorithm is evaluated using synthetic and real world data and shows substantial speedups compared to deterministic HALS. [ABSTRACT FROM AUTHOR]

The progress of technology has been a significant factor in increasing the growth of digital data. Therefore, good data analysis is a necessity for making better decisions. Clustering is one of the most important elements in the field of data analysis. However, the clustering of very large datasets is considered a primary concern. The improvement of computational models along with the ability to cluster huge volumes of data within a reasonable amount of time is thus required. MapReduce is a powerful programming model and an associated implement for processing large datasets with a parallel, distributed algorithm in a computing cluster. In this paper, a MapReduce-based artificial bee colony called MR-ABC is proposed for data clustering. The ABC is implemented based on the MapReduce model in the Hadoop framework and utilized to optimize the assignment of the large data instances to clusters with the objective of minimizing the sum of the squared Euclidean distance between each data instance and the centroid of the cluster to which it belongs. The experimental results demonstrate that our proposed algorithm is well-suited for dealing with massive amounts of data, while the quality level of the clustering results is still maintained. [ABSTRACT FROM AUTHOR]

The growing amount of multimedia documents demanded the efficient knowledge discovery systems. The efficacy of the knowledge discovery systems is influenced by the representation of multimedia documents. The suitable multimedia document representation acts as a platform for multimedia mining tasks. In this paper, a Multimedia Suffix Tree Document model (MSTD) is presented to represent the multimedia documents in a tree based structure. The MSTD model discovers the useful patterns embedded in the multimedia documents and reduces the search time thereby aiding the multimedia mining methods. It provides the complete information of the multimedia documents in one structure. In order to evaluate the proficiency of the proposed MSTD model, the MSTD model based mining methods are proposed. The experiments are conducted with three multimodal multimedia document datasets. The experimental analysis of the proposed methods reveal the significance of MSTD representation for multimedia documents in achieving the significant performance of multimedia mining tasks. [ABSTRACT FROM AUTHOR]

This paper introduces an optimized Windowing based strategy for inducing decision trees in Distributed Data Mining scenarios. Windowing consists in selecting a sample of the available training examples (the window) to induce a decision tree with an usual algorithm, e.g., J48; finding instances not covered by this tree (counter examples) in the remaining training examples, adding them to the window to induce a new tree; and repeating until a termination criterion is met. In this way, the number of training examples required to induce the tree is reduced considerably, while maintaining the expected accuracy levels; which is paid in terms of time performance. Our proposed enhancements solve this by searching for counter examples on GPUs and further reducing their number in the window. The resulting strategy is implemented in JaCa-DDM, our agents & artifacts tool for Distributed Data Mining, keeping the benefits of Windowing, while distributing the process and being faster than the traditional centralized approach, even performing similarly to Bagging and Random Forests in some cases. Experiments in data mining tasks are addressed, including a case study on pixel-based segmentation for the detection of precancerous cervical lesions on medical images. [ABSTRACT FROM AUTHOR]

In many machine learning applications such as computer-aided diagnosis, gene sequence analysis or natural language processing, categorical data appears. For small-scale data set with high dimensions, since relatively small proportion of possible categorical configurations are covered by training samples, conventional methods based on frequency information such as Dirichlet Compound Multinomial distribution usually runs into problems of over-fitting. Latent gaussian process is an effective bayesian non-parametric technique for categorical data modeling, which was proposed as an unsupervised method to embed unlabelled categorical data into a continuous and low-dimensional space through gaussian process. As a probabilistic generative model, latent gaussian process owns the ability of density estimation. In this paper, we propose a generative classification model as a supervised method for labelled categorical data, in which we use latent gaussian process to estimate the class-conditional densities. Since the complexity of gaussian process model can adapt to the size of training data, our method is able to effectively model small-sale categorical data. Experimental results show that our proposal can achieve better classification performance compared with other classification models for categorical data. [ABSTRACT FROM AUTHOR]

This paper reports on a rough set approach to classifying meteorological volumetric radar data used to detect storm events responsible for summer severe weather. The classification of storm cells is a difficult problem due to the complex evolution of storm cells, the high dimensionality of the weather data, and the imprecision and incompleteness of the data. A rough set approach is used to classify different types of meteorological storm events. A considerable of different classification strategies techniques have been considered and compared to determine which approach will best classify the volumetric storm cell data coming from the Radar Decision Support System database of Environment Canada. The criterion for comparison is the accuracy coefficient in the classification over a testing data. The contribution of this paper is a new application of rough set theory in classifying meteorological radar data. ® Elsevier Science B.V. All rights reserved. [Copyright &y& Elsevier]

Early gesture/action recognition is the task of determining the identity of a gesture/action with as few information as possible. Although the topic is relatively new, there are some methods that address this problem. However, existing methods rely on complex modeling procedures, that do not necessarily paid off the computational effort. Thus, simple yet effective and efficient techniques are required for this task. This paper describes a new methodology for early gesture recognition based on the well known naïve Bayes classifier. The method is extremely simple and very fast, yet it compares favorably with more elaborated state of the art methodologies. The naïve baseline is based on three main observations: (1) the effectiveness of the naïve Bayes classifier in text mining problems; (2) the link between natural language processing and computer vision via the bag-of-words representation; and (3) the cumulative-evidence nature of the inference process of naïve Bayes. We evaluated the proposed method in several collections that included segmented and continuous video. Experimental results show that the proposed methodology compares favorably with state of the art methodologies that are more elaborated or were specifically designed for this purpose. [ABSTRACT FROM AUTHOR]

In this paper, we present a novel rotation-invariant and computationally efficient texture descriptor called Dominant Rotated Local Binary Pattern (DRLBP). A rotation invariance is achieved by computing the descriptor with respect to a reference in a local neighborhood. A reference is fast to compute maintaining the computational simplicity of the Local Binary Patterns (LBP). The proposed approach not only retains the complete structural information extracted by LBP, but it also captures the complimentary information by utilizing the magnitude information, thereby achieving more discriminative power. For feature selection, we learn a dictionary of the most frequently occurring patterns from the training images, and discard redundant and non-informative features. To evaluate the performance we conduct experiments on three standard texture datasets: Outex12, Outex 10 and KTH-TIPS. The performance is compared with the state-of-the-art rotation invariant texture descriptors and results show that the proposed method is superior to other approaches. [ABSTRACT FROM AUTHOR]

Outlier detection is an important issue in machine learning and knowledge discovery. The aim is to find the patterns that deviate too much from others. In this paper, we consider constant false-alarm rate (CFAR) outlier detection, and propose a supervised detection method based on normalized residual (NR). For a query point, its NR value related to the training data is compared with a predefined threshold, indicating if it is an outlier. Heretofore, the choice of outlier threshold relied too much on experience, making CFAR detection impossible. We solve the problem by introducing a sufficiently training strategy applying to the given normal instances, gaining a large number of NR values of them, based on which the threshold can be located properly according to the desired false-alarm rate. Theoretical analysis proves that the proposed method can achieve CFAR detection and the most powerful test, regardless of pattern dimension and noise distribution, thus can be widely applied to outlier detection problems. Simulations and real-world data experiments also show that, the proposed method can effectively control the false-alarm rate even when a few training instances are available, and at the same time its operating characteristic is generally better than competing methods. [ABSTRACT FROM AUTHOR]

The massive amount of high-dimensional data in science and engineering demands new trends in data analysis. Subspace techniques have shown remarkable success in numerous problems in computer vision and data mining, where the goal is to recover the low-dimensional structure of data in an ambient space. Traditional subspace methods like PCA and ICA assume that the data is coming from a single manifold. However, the data might come from several (possibly intersected) manifolds (surfaces). This has caused the development of new nonlinear techniques to cluster subspaces of high-dimensional data. In this paper, we propose a new algorithm for subspace clustering of data, where the data consists of several possibly intersected manifolds. To this end, we first propose a curvature constraint to find the shortest path between data points and then use it in Isomap for subspace learning. The algorithm chooses several landmark nodes at random and then checks whether there is a curvature constrained path between each landmark node and all other nodes in the neighborhood graph. It builds a binary feature vector for each point where each entry represents the connectivity of that point to a particular landmark. Then the binary feature vectors could be used as a input of conventional clustering algorithms such as hierarchical clustering. The performed experiments on both synthetic and real data sets confirm the performance of our algorithm. [ABSTRACT FROM AUTHOR]

Pairwise generalized association rules mined from raw data can be used to connect the concepts of multiple ontologies. In this case the items of rules are hierarchically organized and one can use the relations between them in order to reduce rule redundancy. Recently proposed hierarchical interestingness measures address this issue, taking hierarchical information on the antecedent side into account. In this paper, we extend them to the case of considering two hierarchies on both the antecedent and the consequent sides of a rule. The extended measures are then compared with their counterparts as well as with conventional ones. Three real world datasets from the text mining domain with predefined ground truth sets of associations are used for comparison within the framework of instance-based ontology mapping. [ABSTRACT FROM AUTHOR]

Outlier detection is an important problem in data mining that aims to discover useful exceptional and unusual patterns hidden in large data sets. Fraud detection, time series monitoring, intrusion detection and medical condition monitoring are some of the most common applications of outlier detection. Most existing outlier detection methods are based on supervised or unsupervised learning while some others use semi-supervised approaches. However, in many real world applications, there are not enough labeled data for training and only a few positive labeled samples are available. This paper presents an entropy-based solution. The proposed method consists of two phases. First, reliable negative examples are extracted from positive and unlabeled data and then, as the second phase, the entropy-based outlier detection algorithm is employed to detect top N outliers. Many experiments on real and synthetic data sets are performed. The experimental results on synthetic and real data demonstrated superiority of the proposed outlier detection method in comparison of unsupervised state-of-the-art outlier detection strategies. [ABSTRACT FROM AUTHOR]

Abstract: Over recent years the popularity of time series has soared. Given the widespread use of modern information technology, a large number of time series may be collected. As a consequence there has been a dramatic increase in the amount of interest in querying and mining such data. A vital component in many types of time series analyses is the choice of an appropriate dissimilarity measure. Numerous measures have been proposed to date, with the most successful ones based on dynamic programming. One of such measures is longest common subsequence (LCSS). In this paper, we propose a parametrical extension of LCSS based on derivatives. In contrast to well-known measures from the literature, our approach considers the general shape of a time series rather than point-to-point function comparison. The new dissimilarity measure is used in classification with the nearest neighbor rule. In order to provide a comprehensive comparison, we conducted a set of experiments, testing effectiveness on 47 real time series. Experiments show that our method provides a higher quality of classification compared with LCSS on examined data sets. [Copyright &y& Elsevier]

Abstract: It is essential to extract the discriminative information for similar handwritten Chinese character recognition (SHCCR) that plays a key role to improve the performance of handwritten Chinese character recognition. This paper first introduces a new manifold learning based subspace learning algorithm, discriminative locality alignment (DLA), to SHCCR. Afterward, we propose the kernel version of DLA, kernel discriminative locality alignment (KDLA), and carefully prove that learning KDLA is equal to conducting kernel principal component analysis (KPCA) followed by DLA. This theoretical investigation can be utilized to better understand KDLA, i.e., the subspace spanned by KDLA is essentially the subspace spanned by DLA on the principal components of KPCA. Experimental results demonstrate that DLA and KDLA are more effective than representative discriminative information extraction algorithms in terms of recognition accuracy. [Copyright &y& Elsevier]

Abstract: High level visual cognitive abilities such as scene understanding and behavioural analysis are modulated by attentive selective processes. These in turn rely on pre-attentive operations delivering perceptual organisation of the visual input and enabling the extraction of meaningful “chunks” of information. Specifically, the extraction and prioritisation of moving objects is a crucial step in the processing of dynamic scenes. Motion is of course a powerful cue for grouping regions and segregating objects but not all kinds of motion are equally meaningful and should be equally attended. On a coarse level, most interesting moving objects are associated with coherent motion, reflecting our sensitivity to biological motion. On the other hand, attention operates on a higher level, prioritising what moves differently with respect to both its surrounding and the global scene. In this paper, we propose how a qualitative segmentation of multiscale spatiotemporal energy features according to their frequency spectrum distribution can be used to pre-attentively extract regions of interest. We also show that discrimination boundaries between classes in the segmentation phase can be learned in an automatic and efficient way by a Support Vector Machine classifier in a multi-class implementation. Motion-related features are shown to best predict human fixations on an extensive dataset. The model generalises well to datasets other than that used for training, if scale is taken into account in the feature extraction step. Regions labelled as coherently moving are clustered in moving object files, described by the magnitude and phase of the pooled motion energy. The method succeeds in extracting meaningful moving objects from the background and identifying other less interesting motion patterns. A saliency function is finally computed on an object basis, instead of on a pixel basis, as in most current approaches. The same features are thus used for segmentation and selective attention and can be further used for recognition and scene interpretation. [Copyright &y& Elsevier]

Abstract: Feature selection methods have been extensively applied in machine learning tasks, such as computer vision, pattern recognition, and data mining. These methods aim to identify a subset of the original features with high discriminating power. Among them, the feature selection technique for unsupervised tasks is more attractive since the cost to obtain the labels of the data and/or the information between classes is often high. On the other hand, the low-dimensional manifold of the “same” class data is usually revealed by considering the local invariance of the data structure, it may not be adequate to deal with unsupervised tasks where the class information is completely absent. In this paper, a novel feature selection method, called Minimum–maximum local structure information Laplacian Score (MMLS), is proposed to minimize the within-locality information (i.e., preserving the manifold structure of the “same” class data) and to maximize the between-locality information (i.e., maximizing the information between the manifold structures of the “different” class data) at the same time. The effectiveness of the proposed algorithm is demonstrated with experiments on classification and clustering. [Copyright &y& Elsevier]

Abstract: Class imbalance and class overlap are two of the major problems in data mining and machine learning. Several studies have shown that these data complexities may affect the performance or behavior of artificial neural networks. Strategies proposed to face with both challenges have been separately applied. In this paper, we introduce a hybrid method for handling both class imbalance and class overlap simultaneously in multi-class learning problems. Experimental results on five remote sensing data show that the combined approach is a promising method. [Copyright &y& Elsevier]

Abstract: Similarity is the core problem of clustering. Clustering algorithms that are based on a certain, fixed type of similarity are not sufficient to explore complicated structures. In this paper, a constructing method for multiple similarity is proposed to deal with complicated structures of data sets. Multiple similarity derives from the local modification of the initial similarity, based on the feedback information of elementary clusters. Combined with the proposed algorithm, the repeated modifications of local similarity measurement generate a hierarchical clustering result. Some synthetic and real data sets are employed to exhibit the superiority of the new clustering algorithm. [Copyright &y& Elsevier]

Abstract: Data clustering is an important technique in data mining. It is a method of partitioning data into clusters, in which each cluster must have data of great similarity and different clusters must have data of high dissimilarity. A lot of clustering algorithms are found in the literature. In general, there is no single algorithm that is suitable for all types of data, conditions and applications. Each algorithm has its own advantages, limitations and shortcomings. Therefore, introducing novel and effective approaches for data clustering is an open and active research area. This paper presents a novel binary search algorithm for data clustering that not only finds high quality clusters but also converges to the same solution in different runs. In the proposed algorithm a set of initial centroids are chosen from different parts of the test dataset and then optimal locations for the centroids are found by thoroughly exploring around of the initial centroids. The simulation results using six benchmark datasets from the UCI Machine Learning Repository indicate that proposed algorithm can efficiently be used for data clustering. [Copyright &y& Elsevier]

Abstract: In this paper, the relationship between a selected subset of attribute set of a decision system via feature selection by an optimal algorithm and a reduct of attribute set under the meaning of Pawlak’s rough set is discussed. This selected subset is considered as a solution of the optimal algorithm. It is verified that a locally optimal solution is surely not a reduct while a reduct must be a globally optimal solution. Based on these assertions, a new optimal algorithm, called blindly deleting algorithm with an inverse ordering (BDAIO), is proposed to find a real reduct of a decision information system by remedying the selected attribute subset. Several standard data sets from UCI repository are implemented showing validity of the proposal. [Copyright &y& Elsevier]

Abstract: Fractal theory presents a large number of applications to image and signal analysis. Although the fractal dimension can be used as an image object descriptor, a multiscale approach, such as multiscale fractal dimension (MFD), increases the amount of information extracted from an object. MFD provides a curve which describes object complexity along the scale. However, this curve presents much redundant information, which could be discarded without loss in performance. Thus, it is necessary the use of a descriptor technique to analyze this curve and also to reduce the dimensionality of these data by selecting its meaningful descriptors. This paper shows a comparative study among different techniques for MFD descriptors generation. It compares the use of well-known and state-of-the-art descriptors, such as Fourier, Wavelet, Polynomial Approximation (PA), Functional Data Analysis (FDA), Principal Component Analysis (PCA), Symbolic Aggregate Approximation (SAX), kernel PCA, Independent Component Analysis (ICA), geometrical and statistical features. The descriptors are evaluated in a classification experiment using Linear Discriminant Analysis over the descriptors computed from MFD curves from two data sets: generic shapes and rotated fish contours. Results indicate that PCA, FDA, PA and Wavelet Approximation provide the best MFD descriptors for recognition and classification tasks. [Copyright &y& Elsevier]

Abstract: This paper addresses the problem of proportional data modeling and clustering using mixture models, a problem of great interest and of importance for many practical pattern recognition, image processing, data mining and computer vision applications. Finite mixture models are broadly applicable to clustering problems. But, they involve the challenging problem of the selection of the number of clusters which requires a certain trade-off. The number of clusters must be sufficient to provide the discriminating capability between clusters required for a given application. Indeed, if too many clusters are employed overfitting problems may occur and if few are used we have a problem of underfitting. Here we approach the problem of modeling and clustering proportional data using infinite mixtures which have been shown to be an efficient alternative to finite mixtures by overcoming the concern regarding the selection of the optimal number of mixture components. In particular, we propose and discuss the consideration of infinite Liouville mixture model whose parameter values are fitted to the data through a principled Bayesian algorithm that we have developed and which allows uncertainty in the number of mixture components. Our experimental evaluation involves two challenging applications namely text classification and texture discrimination, and suggests that the proposed approach can be an excellent choice for proportional data modeling. [Copyright &y& Elsevier]

Abstract: The present paper considers the problem of partitioning a dataset into a known number of clusters using the sum of squared errors criterion (SSE). A new clustering method, called DE-KM, which combines differential evolution algorithm (DE) with the well known K-means procedure is described. In the method, the K-means algorithm is used to fine-tune each candidate solution obtained by mutation and crossover operators of DE. Additionally, a reordering procedure which allows the evolutionary algorithm to tackle the redundant representation problem is proposed. The performance of the DE-KM clustering method is compared to the performance of differential evolution, global K-means method, genetic K-means algorithm and two variants of the K-means algorithm. The experimental results show that if the number of clusters K is sufficiently large, DE-KM obtains solutions with lower SSE values than the other five algorithms. [Copyright &y& Elsevier]

Abstract: Benchmarking pattern recognition, machine learning and data mining methods commonly relies on real-world data sets. However, there are some disadvantages in using real-world data. On one hand collecting real-world data can become difficult or impossible for various reasons, on the other hand real-world variables are hard to control, even in the problem domain; in the feature domain, where most statistical learning methods operate, exercising control is even more difficult and hence rarely attempted. This is at odds with the scientific experimentation guidelines mandating the use of as directly controllable and as directly observable variables as possible. Because of this, synthetic data possesses certain advantages over real-world data sets. In this paper we propose a method that produces synthetic data with guaranteed global and class-specific statistical properties. This method is based on overlapping class densities placed on the corners of a regular k-simplex. This generator can be used for algorithm testing and fair performance evaluation of statistical learning methods. Because of the strong properties of this generator researchers can reproduce each others experiments by knowing the parameters used, instead of transmitting large data sets. [Copyright &y& Elsevier]