Competition Dataset Research Articles

Systematic Approach for Detecting Text in Images Using Supervised Learning

Locating text data in images automatically has been a challenging task. In this approach, we build a three stage system for text detection purpose. This system utilizes tensor voting and Completed Local Binary Pattern (CLBP) to classify text and non-text regions. While tensor voting generates the text line information, which is very useful for localizing candidate text regions, the Nearest Neighbor classifier trained on discriminative features obtained by the CLBP-based operator is used to refine the results. The whole algorithm is implemented in MATLAB and applied to all images of ICDAR 2011 Robust Reading Competition data set. Experiments show the promising performance of this method.

Scene Text Detection via Connected Component Clustering and Nontext Filtering

In this paper, we present a new scene text detection algorithm based on two machine learning classifiers: one allows us to generate candidate word regions and the other filters out nontext ones. To be precise, we extract connected components (CCs) in images by using the maximally stable extremal region algorithm. These extracted CCs are partitioned into clusters so that we can generate candidate regions. Unlike conventional methods relying on heuristic rules in clustering, we train an AdaBoost classifier that determines the adjacency relationship and cluster CCs by using their pairwise relations. Then we normalize candidate word regions and determine whether each region contains text or not. Since the scale, skew, and color of each candidate can be estimated from CCs, we develop a text/nontext classifier for normalized images. This classifier is based on multilayer perceptrons and we can control recall and precision rates with a single free parameter. Finally, we extend our approach to exploit multichannel information. Experimental results on ICDAR 2005 and 2011 robust reading competition datasets show that our method yields the state-of-the-art performance both in speed and accuracy.

Breaking Out of the Coup Trap

This article systematically assesses whether open political competition ends the coup trap. We use an original data set of coup reports, electoral competition, and socioeconomic data spanning the 20th century in 18 Latin American countries. Our models, which are robust to multiple-comparison tests, generate support for a central claim: While recent experience with military coups increases the risk of incumbents being overthrown, the establishment of open political competition ends cycles of political instability. The risk of being overthrown consequently declines with time for presidents in noncompetitive systems, but remains stable—and, on average, substantially lower—in competitive polities. Coups do not vary with levels of development or of inequality, economic growth rates, or the legislative powers of the presidency.

Designing discriminative spatial filter vectors in motor imagery brain–computer interface

The problem of a volume conduction effect in electroencephalography is considered one of the challenging issues in brain–computer interface (BCI) community. In this article, we propose a novel method of designing a class‐discriminative spatial filter assuming that a combination of spatial pattern vectors, irrespective of the eigenvalues of the common spatial pattern (CSP), can produce better performance in terms of classification accuracy. We select discriminative spatial filter vectors that determine features in a pairwise manner, that is, eigenvectors of the K largest eigenvalue and the K smallest eigenvalue. Although the pair of the eigenvectors of the K largest and the K smallest eigenvalues helps extract discriminative features, we believe that a different set of eigenvector pairs is more appropriate to extract class‐discriminative features. In our experimental results using the publicly available dataset of BCI Competition IV, we show that the proposed method outperformed the conventional CSP methods and a filter‐bank CSP. © 2013 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 23, 147–151, 2013

Classification of covariance matrices using a Riemannian-based kernel for BCI applications

The use of spatial covariance matrix as a feature is investigated for motor imagery EEG-based classification in brain–computer interface applications. A new kernel is derived by establishing a connection with the Riemannian geometry of symmetric positive definite matrices. Different kernels are tested, in combination with support vector machines, on a past BCI competition dataset. We demonstrate that this new approach outperforms significantly state of the art results, effectively replacing the traditional spatial filtering approach.

Scaling factorization machines to relational data

The most common approach in predictive modeling is to describe cases with feature vectors (aka design matrix). Many machine learning methods such as linear regression or support vector machines rely on this representation. However, when the underlying data has strong relational patterns, especially relations with high cardinality, the design matrix can get very large which can make learning and prediction slow or even infeasible.This work solves this issue by making use of repeating patterns in the design matrix which stem from the underlying relational structure of the data. It is shown how coordinate descent learning and Bayesian Markov Chain Monte Carlo inference can be scaled for linear regression and factorization machine models. Empirically, it is shown on two large scale and very competitive datasets (Netflix prize, KDDCup 2012), that (1) standard learning algorithms based on the design matrix representation cannot scale to relational predictor variables, (2) the proposed new algorithms scale and (3) the predictive quality of the proposed generic feature-based approach is as good as the best specialized models that have been tailored to the respective tasks.

Domain transformation approach to deterministic optimization of examination timetables

In this paper we introduce a new optimization method for the examinations scheduling problem. Rather than attempting direct optimization of assignments of exams to specific time-slots, we perform permutations of slots and reassignments of exams upon the feasible (but not optimal) schedules obtained by the standard graph colouring method with Largest Degree ordering. The proposed optimization methods have been evaluated on the University of Toronto, University of Nottingham and International Timetabling Competition (ITC2007) datasets. It is shown that the proposed method delivers competitive results compared to other constructive methods in the timetabling literature on both the Nottingham and Toronto datasets, and it maintains the same optimization pattern of the solution improvement on the ITC2007 dataset. A deterministic pattern obtained for all benchmark datasets, makes the proposed method more understandable to the users.

Hybrid biogeography based simultaneous feature selection and MHC class I peptide binding prediction using support vector machines and random forests

Accurate detection of peptides binding to specific Major Histocompatibility Complex Class I (MHC-I) molecules is extremely important for understanding the underlying process of the immune system, as well as for effective vaccine design and developing immunotherapies. Development of learning algorithms and their application for binding predictions have thus speeded up the state-of-the-art in immunological research, in a cost-effective manner. In this work, we propose the application of a hybrid filter-wrapper algorithm employing concepts from the recently developed biogeography based optimization algorithm, in conjunction with SVM and Random Forests for identification of MHC-I binding peptides. In the process, we demonstrate the effectiveness of this evolutionary technique, coupled with weighted heuristics, for the construction of improved prediction models. The experiments have been carried out for the CoEPrA competition datasets (accessible online at: http://www.coepra.org) and the results show a marked improvement over the winner results in some situations and comparably good with regard to others .We thus hope to initiate further research on the application of this new bio-inspired methodology for immunological research.

Scene text detection using graph model built upon maximally stable extremal regions

Scene text detection could be formulated as a bi-label (text and non-text regions) segmentation problem. However, due to the high degree of intraclass variation of scene characters as well as the limited number of training samples, single information source or classifier is not enough to segment text from non-text background. Thus, in this paper, we propose a novel scene text detection approach using graph model built upon Maximally Stable Extremal Regions (MSERs) to incorporate various information sources into one framework. Concretely, after detecting MSERs in the original image, an irregular graph whose nodes are MSERs, is constructed to label MSERs as text regions or non-text ones. Carefully designed features contribute to the unary potential to assess the individual penalties for labeling a MSER node as text or non-text, and color and geometric features are used to define the pairwise potential to punish the likely discontinuities. By minimizing the cost function via graph cut algorithm, different information carried by the cost function could be optimally balanced to get the final MSERs labeling result. The proposed method is naturally context-relevant and scale-insensitive. Experimental results on the ICDAR 2011 competition dataset show that the proposed approach outperforms state-of-the-art methods both in recall and precision.

De-identification Methods for Open Health Data: The Case of the Heritage Health Prize Claims Dataset

BackgroundThere are many benefits to open datasets. However, privacy concerns have hampered the widespread creation of open health data. There is a dearth of documented methods and case studies for the creation of public-use health data. We describe a new methodology for creating a longitudinal public health dataset in the context of the Heritage Health Prize (HHP). The HHP is a global data mining competition to predict, by using claims data, the number of days patients will be hospitalized in a subsequent year. The winner will be the team or individual with the most accurate model past a threshold accuracy, and will receive a US $3 million cash prize. HHP began on April 4, 2011, and ends on April 3, 2013.ObjectiveTo de-identify the claims data used in the HHP competition and ensure that it meets the requirements in the US Health Insurance Portability and Accountability Act (HIPAA) Privacy Rule.MethodsWe defined a threshold risk consistent with the HIPAA Privacy Rule Safe Harbor standard for disclosing the competition dataset. Three plausible re-identification attacks that can be executed on these data were identified. For each attack the re-identification probability was evaluated. If it was deemed too high then a new de-identification algorithm was applied to reduce the risk to an acceptable level. We performed an actual evaluation of re-identification risk using simulated attacks and matching experiments to confirm the results of the de-identification and to test sensitivity to assumptions. The main metric used to evaluate re-identification risk was the probability that a record in the HHP data can be re-identified given an attempted attack.ResultsAn evaluation of the de-identified dataset estimated that the probability of re-identifying an individual was .0084, below the .05 probability threshold specified for the competition. The risk was robust to violations of our initial assumptions.ConclusionsIt was possible to ensure that the probability of re-identification for a large longitudinal dataset was acceptably low when it was released for a global user community in support of an analytics competition. This is an example of, and methodology for, achieving open data principles for longitudinal health data.

On the use of multi neighbourhood structures within a Tabu-based memetic approach to university timetabling problems

Finding a good university timetabling system is not a simple task for a higher educational organisation. As a result, many approaches to generating sufficiently good solutions have been introduced. This is mainly due to the high complexity within the search landscape; moreover, each educational organisation has its own rules and specifications. In this paper, a Tabu-based memetic algorithm that hybridises a genetic algorithm with a Tabu Search algorithm is proposed as an improved algorithm for university timetabling problems. This algorithm is employed on a set of neighbourhood structures during the search process with the aim of gaining significant improvements in solution quality. The sequence of neighbourhood structures has been considered to understand its effect on the search space. Random, best and general sequences of neighbourhood structures have been evaluated in this work. The concept of a Tabu list is embedded to control the selection of neighbourhood structures that are not dependent on the problem domains during the optimisation process after the crossover and mutation operators are applied to the selected solutions from the population pool. The algorithm will penalise neighbourhood structures that are unable to generate better solutions. The proposed algorithm has been applied and evaluated against the latest methodologies in the literature with respect to standard benchmark problems. We demonstrate that the proposed algorithm produces some of the best known results when tested on ITC2007 competition datasets.

Learning discriminative patterns for self-paced EEG-based motor imagery detection

Detecting motor imagery activities versus non-control in brain signals is the basis of self-paced brain-computer interfaces (BCIs), but also poses a considerable challenge to signal processing due to the complex and non-stationary characteristics of motor imagery as well as non-control. This paper presents a self-paced BCI based on a robust learning mechanism that extracts and selects spatio-spectral features for differentiating multiple EEG classes. It also employs a non-linear regression and post-processing technique for predicting the time-series of class labels from the spatio-spectral features. The method was validated in the BCI Competition IV on Dataset I where it produced the lowest prediction error of class labels continuously. This report also presents and discusses analysis of the method using the competition data set.

ADHD diagnosis from multiple data sources with batch effects

The Attention Deficit Hyperactivity Disorder (ADHD) affects the school-age population and has large social costs. The scientific community is still lacking a pathophysiological model of the disorder and there are no objective biomarkers to support the diagnosis. In 2011 the ADHD-200 Consortium provided a rich, heterogeneous neuroimaging dataset aimed at studying neural correlates of ADHD and to promote the development of systems for automated diagnosis. Concurrently a competition was set up with the goal of addressing the wide range of different types of data for the accurate prediction of the presence of ADHD. Phenotypic information, structural magnetic resonance imaging (MRI) scans and resting state fMRI recordings were provided for nearly 1000 typical and non-typical young individuals. Data were collected by eight different research centers in the consortium. This work is not concerned with the main task of the contest, i.e., achieving a high prediction accuracy on the competition dataset, but we rather address the proper handling of such a heterogeneous dataset when performing classification-based analysis. Our interest lies in the clustered structure of the data causing the so-called batch effects which have strong impact when assessing the performance of classifiers built on the ADHD-200 dataset. We propose a method to eliminate the biases introduced by such batch effects. Its application on the ADHD-200 dataset generates such a significant drop in prediction accuracy that most of the conclusions from a standard analysis had to be revised. In addition we propose to adopt the dissimilarity representation to set up effective representation spaces for the heterogeneous ADHD-200 dataset. Moreover we propose to evaluate the quality of predictions through a recently proposed test of independence in order to cope with the unbalancedness of the dataset.

Simple instance selection for bankruptcy prediction

Instance selection or outlier detection is an important task during data mining, which focuses on filtering out bad data from a given dataset. However, there is no rigid mathematical definition of what constitutes an outlier and an outlier is not a binary property. Therefore, different volumes of outliers may be detected depending on the setting of the threshold for what constitutes an outlier, e.g., the distance in distance-based outlier detection. In this study, we examine bankruptcy prediction performance achieved after removal of different outlier volumes from four widely used datasets, namely the Australian, German, Japanese, and UC Competition datasets. Specifically, a simple distance-based clustering outlier detection method is used. In addition, four popular classification techniques are compared, artificial neural networks, decision trees, logistic regression, and support vector machines. Experiments are conducted to examine (1) the prediction performance of the bankruptcy prediction models with and without instance selection, (2) the stability of bankruptcy prediction models after the removal of outliers from the testing set, and (3) the characteristics of these four different datasets. The results show that with the German dataset it is much more difficult for the prediction models to provide high rates of accuracy after outlier removal, while it is easier with the UC Competition dataset. Removing 50% of the outliers can lead to optimal performance of these four models. In addition, using the removed outliers to test the prediction accuracy of these models, we find that it is support vector machines (SVM) that provide the highest rate of prediction accuracy and perform with much more stability and good noise tolerance than the other three prediction models. Furthermore, the prediction accuracy of the SVM model followed by instance selection is similar to the one without instance selection (i.e., the SVM baseline). In other words, the difference in performance between the SVM and the SVM baseline is the least of the three models in comparison with their corresponding baselines.

Subject and class specific frequency bands selection for multiclass motor imagery classification

AbstractEEG‐based discrimination among motor imagery states has been widely studied for brain‐computer interfaces (BCIs) due to the great potential for real‐life applications. However, in terms of designing a motor imagery‐based BCI system, a lot of research in the literature either uses a frequency band of interest selected manually based on the visual analysis of EEG data or is set to a general broad band, causing performance degradation in classification. In this article, we propose a novel method of selecting subject and class specific frequency bands based on the analysis of a channel‐frequency matrix, which we call a channel‐frequency map. We operate the classification process for each frequency band individually, i.e., spatial filtering, feature extraction, and classification, and determine a class label for an input EEG by considering the outputs from multiple classifiers together at the end. From our experiments on a public dataset of BCI Competition IV (2008) II‐a that includes four motor imagery tasks from nine subjects, the proposed algorithm outperformed the common spatial pattern (CSP) algorithm in a broad band and a filter bank CSP algorithm on average in terms of cross‐validation and session‐to‐session transfer rate. Furthermore, a considerable increase of classification accuracy has been achieved for certain subjects. We also would like to note that the proposed data‐driven frequency bands selection method is applicable to other kinds of single‐trial EEG classifications that are based on modulations of brain rhythms, by no means limited to motor imagery‐based BCI applications. © 2011 Wiley Periodicals, Inc. Int J Imaging Syst Technol, 21, 123–130, 2011

Classifying Single Trail Electroencephalogram Using Gaussian Smoothened Fast Hartley Transform for Brain Computer Interface during Motor Imagery

Problem statement: Brain-Computer Interface (BCI) is a emerging research area which translates the brain signals for any motor related actions into computer understandable signals by capturing the signal, processing the signal and classifying the motor imagery. This area of work finds various applications in neuroprosthetics. Mental activity leads to changes of electrophysiological signals like the Electroencephalogram (EEG) or Electrocorticogram (ECoG). Approach: The BCI system detects such changes and transforms it into a control signal which can, for example, be used as to control a electric wheel. In this study the BCI paradigm is tested by our proposed Gaussian smoothened Fast Hartley Transform (GS-FHT) which is used to compute the energies of different motor imageries the subject thinks after selecting the required frequencies using band pass filter. Results: We apply this procedure to BCI Competition dataset IVA, a publicly available EEG repository. Conclusion: The evaluations of preprocessed signals showed that the extracted features were interpretable and can lead to high classification accuracy by various mining algorithms.

Disaggregation & aggregation of time series components: A hybrid forecasting approach using generalized regression neural networks and the theta method

A hybrid forecasting method is proposed which leverages from statistical and neural network techniques to perform multi-step ahead forecasting. The proposed method is based on the disaggregation of time series components, the prediction of each component individually and the reassembling of the extrapolations to obtain an estimation for the global data. The STL decomposition procedure from the literature [ 5] is implemented to obtain the seasonal, trend and irregular components of the time series, whilst Generalized Regression Neural Networks (GRNN) [12] are used to perform out-of sample extrapolations of the seasonal and residual components. The univariate Theta model is employed for the estimation of the directional component. The application of the GRNN is based on the dynamic calibration of the training process for each of the seasonal and irregular components individually. The proposed hybrid forecasting method is applied to 60 time series from the NN3 competition and 227 time series from the M1 Competition dataset, to obtain 18 out-of sample predictions. The results from the application demonstrate that the proposed method can outperform standard statistical techniques in the literature. One of the main contributions of the current research lies in the investigation of the strengths and weaknesses of the GRNN in extrapolating structural components of time series.

Regularizing Common Spatial Patterns to Improve BCI Designs: Unified Theory and New Algorithms

One of the most popular feature extraction algorithms for brain-computer interfaces (BCI) is common spatial patterns (CSPs). Despite its known efficiency and widespread use, CSP is also known to be very sensitive to noise and prone to overfitting. To address this issue, it has been recently proposed to regularize CSP. In this paper, we present a simple and unifying theoretical framework to design such a regularized CSP (RCSP). We then present a review of existing RCSP algorithms and describe how to cast them in this framework. We also propose four new RCSP algorithms. Finally, we compare the performances of 11 different RCSP (including the four new ones and the original CSP), on electroencephalography data from 17 subjects, from BCI competition datasets. Results showed that the best RCSP methods can outperform CSP by nearly 10% in median classification accuracy and lead to more neurophysiologically relevant spatial filters. They also enable us to perform efficient subject-to-subject transfer. Overall, the best RCSP algorithms were CSP with Tikhonov regularization and weighted Tikhonov regularization, both proposed in this paper.

A Hybrid Approach to Detect and Localize Texts in Natural Scene Images

Text detection and localization in natural scene images is important for content-based image analysis. This problem is challenging due to the complex background, the non-uniform illumination, the variations of text font, size and line orientation. In this paper, we present a hybrid approach to robustly detect and localize texts in natural scene images. A text region detector is designed to estimate the text existing confidence and scale information in image pyramid, which help segment candidate text components by local binarization. To efficiently filter out the non-text components, a conditional random field (CRF) model considering unary component properties and binary contextual component relationships with supervised parameter learning is proposed. Finally, text components are grouped into text lines/words with a learning-based energy minimization method. Since all the three stages are learning-based, there are very few parameters requiring manual tuning. Experimental results evaluated on the ICDAR 2005 competition dataset show that our approach yields higher precision and recall performance compared with state-of-the-art methods. We also evaluated our approach on a multilingual image dataset with promising results.

Comparison of Pre-Processing and Classification Techniques for Single-Trial and Multi-Trial P300-Based Brain Computer Interfaces

The P300 component of Event Related Brain Potentials (ERP) is commonly used in Brain Computer Interfaces (BCI) to translate the intentio ns of an individual into commands for external devices. The P300 response, however, resides in a s ignal environment of high background noise. Consequently, the main problem in developing a P300-based BCI lies in identifying the P300 response in the presence of this noise. Traditionally, atten uating the background activity of P300 data is done by averaging multiple trials of recorded signals. This method, though effective, suffers two drawbacks. First, collecting multiple trials of data is time c onsuming and delays the BCI response. Second, latency distortions may appear in the averaged result due t o variable time-locking of the P300 in the individu al trials. Problem statement: The use of single-trial P300 data overcomes both t hese shortcomings. However, single-trial data must be properly denoise d to allow for reliable BCI operation. Single-trial P300-based BCIs have been implemented using a variety of signal processing techniques and classification methodologies. However, comparing the accuracies of these systems to other multi-trial systems is likely to include the comparison of more than just the trial format (single-trial/multi-tri al) as the data quality and recording circumstances are li kely to be dissimilar. Approach: This issue was directly addressed by comparing the performance comparison of three different preprocessing agents and three classification methodologies on the same data set over both the single-trial and multi-trial settings. The P300 data set of BCI Competition II w as used to facilitate this comparison. Results: The LDA classifier exhibited the best performance in cl assifying unseen P300 spatiotemporal features in both the single-trial (74.19%) and multi-trial form at (100%). It is also very efficient in terms of computational and memory requirements. Conclusion: This study can serve as a general guide for practitioners developing single-trial and multi-tri al P300-based BCI systems, particularly for selecti ng appropriate pre-processing agents and classificatio n methodologies for inclusion. The possibilities fo r future study include the investigation of double-tr ial and triple-trial P300 system based on the LDA classifier. The time savings of such approaches wil l still be significant. It is very likely that such systems would benefit from accuracies higher than t he one obtained in this study for single-trial LDA (74.19%).