Specific Problem Domain Research Articles

Towards Continuous Quality Control for Spoken Language Corpora


 This paper describes the development of a systematic approach to the creation, management and curation of linguistic resources, particularly spoken language corpora. It also presents first steps towards a framework for continuous quality control to be used within external research projects by non-technical users, and discuss various domain and discipline specific problems and individual solutions. The creation of spoken language corpora is not only a time-consuming and costly process, but the created resources often represent intangible cultural heritage, containing recordings of, for example, extinct languages or historical events. Since high quality resources are needed to enable re-use in as many future contexts as possible, researchers need to be provided with the necessary means for quality control. We believe that this includes methods and tools adapted to Humanities researchers as non-technical users, and that these methods and tools need to be developed to support existing tasks and goals of research projects.

Massively Parallel Rule-Based Interpreter Execution on GPUs Using Thread Compaction

Interpreters are well researched in the field of compiler construction and program generation. They are typically used to realize program execution of different programming languages without a compilation step. However, they can also be used to model complex rule-based simulations: The interpreter applies all rules one after another. These can be iteratively applied on a globally updated state in order to get the final simulation result. Many simulations for domain-specific problems already leverage the parallel processing capabilities of Graphics Processing Units (GPUs). They use hardware-specific tuned rule implementations to achieve maximum performance. However, every interpreter-based system requires a high-level algorithm that detects active rules and determines when they are evaluated. A common approach in this context is the use of different interpreter routines for every problem domain. Executing such functions in an efficient way mainly involves dealing with hardware peculiarities like thread divergences, ALU computations and memory operations. Furthermore, the interpreter is often executed on multiple states in parallel these days. This is particularly important for heuristic search or what-if analyses, for instance. In this paper, we present a novel and easy-to-implement method based on thread compaction to realize generic rule-based interpreters in an efficient way on GPUs. It is optimized for many states using a specially designed memory layout. Benchmarks on our evaluation scenarios show that the performance can be significantly increased in comparison to existing commonly-used implementations.

Virtualization Based Efficient Service Matching and Discovery in Internet of Things

Internet of Things (IoT) is getting more popular day by day, which triggers its adoption for solving domain specific problems. Cities are becoming smart by gathering the context knowledge through sensors and controlling specific parameters through actuators. Dynamically discovering and integrating different data streams from different sensors is a major challenge these days. In this paper, a service matchmaking algorithm is presented for service discovery utilizing IoT devices and services in a particular geographic area. It helps us to identify services based on a variety of parameters (location, query size and processing time, etc.). Customization of service selection and discovery are also explored. The conceptual framework is provided for the proposed model along with a matchmaking algorithm based on IoT devices virtualization. The simulation results elaborate the increased complexity of processing time with respect to the increasing pool of available services. The average processing time varies as the number of conditions are multiplied. Query size and complexity increases with additional number of filters and conditions which results in the reduction of the number of matching services. Moreover, upon decreasing the radius of geographic search area, the number of candidate services decreases for service matching algorithm. This is based on the assumption that IoT devices and services are evenly distributed in a given geographic area. Similarly, the remaining energy of IoT devices is also assumed to be uniformly distributed and, therefore, if we are interested in IoT devices or services with more residual energy, then a limited number of IoT devices or services will fulfill this criterion.

Evaluation of Weakly Supervised Learning Paradigms on Automatic Visual Inspection

In pattern recognition supervised learning algorithms has been used to model the relationships among the characteristics of a group of objects (or patterns) and their class labels. This model is build based on a set of training images that have been labelled by an expert in a specific problem domain. In general, the labelling process consists in defining the object's class label as well as demarcating the precise part of the image in which it is located the object of interest. In many computer vision applications, for example in automatic visual inspection, the labelling process may be a laborious task which requires an extensive work, it could even become impractical when a large number of training images are needed to train the pattern recognition system. This limitation has led to the development of new learning algorithms that allow some ambiguity or weakness in the way class labels are assigned. These algorithms are known as weakly supervised learning algorithms. In this paper we present an evaluation of different weakly supervised learning paradigms grouped according to the representation of the object of interest (in one or several characteristic vectors) associated to its class label (single or multiple label). We perform different experiments in the automatic visual inspection context, using a benchmark dataset of texture images with artificial defects. We test different weakly supervised learning algorithms. Results show that representing an object using multiple instances, presents better results in the identification of defective texture image

Using off-the-shelf data-human interface platforms: traps and tricks

With the development of learning algorithms, the constantly increasing computing power and the available amount of multimedia data, the adoption rate of data science techniques is steadily growing. Machine and deep learning algorithms are already used in a wide variety of ways to solve domain-specific problems. However, the potential of such methodologies will be fulfilled when also non specialized data scientists will be empowered with their use. Focusing on such perspective, this work does not deal with a classical data science problem, but instead exploits existing and available easy to use data-human interfaces. To this aim, we picked an exemplar scenario, amounting to an existing qualitative activity recognition data set that was in the past analyzed utilizing feature selection techniques and custom machine learning paradigms. We here verify how it is today possible, without changing the default settings and/or performing any type of feature selection, to employ the machine and deep learning algorithms provided by different publicly accessible tools (namely, Weka, Orange, Ludwig and KNIME) to address the same problem. Nevertheless, not all of the utilized platforms and algorithms provided satisfactorily results: we here finally discuss the possible issues and opportunities posed by such approach.

Selecting a diverse set of benchmark instances from a tunable model problem for black-box discrete optimization algorithms

As the number of practical applications of discrete black-box metaheuristics is growing faster and faster, the benchmarking of these algorithms is rapidly gaining importance. While new algorithms are often introduced for specific problem domains, researchers are also interested in which general problem characteristics are hard for which type of algorithm. The W-Model is a benchmark function for discrete black-box optimization, which allows for the easy, fast, and reproducible generation of problem instances exhibiting characteristics such as ruggedness, deceptiveness, epistasis, and neutrality in a tunable way. We conduct the first large-scale study with the W-Model in its fixed-length single-objective form, investigating 17 algorithm configurations (including Evolutionary Algorithms and local searches) and 8372 problem instances. We develop and apply a machine learning methodology to automatically discover several clusters of optimization process runtime behaviors as well as their reasons grounded in the algorithm and model parameters. Both a detailed statistical evaluation and our methodology confirm that the different model parameters allow us to generate problem instances of different hardness, but also find that the investigated algorithms struggle with different problem characteristics. With our methodology, we select a set of 19 diverse problem instances with which researchers can conduct a fast but still in-depth analysis of algorithm performance. The best-performing algorithms in our experiment were Evolutionary Algorithms applying Frequency Fitness Assignment, which turned out to be robust over a wide range of problem settings and solved more instances than the other tested algorithms.

Maturity Models Architecture: A large systematic mapping

Maturity models are widespread in research and in particular, IT practitioner communities. However, theoretically sound, methodologically rigorous and empirically validated maturity models are quite rare. This systematic mapping paper focuses on the challenges faced during the development of maturity models. More specifically, it explores the literature on maturity models and standard guidelines to develop maturity models, the challenges identified and solutions proposed. Our systematic mapping revealed over six hundred articles on maturity models. Extant literature reveals that researchers have primarily focused on developing new maturity models pertaining to domain-specific problems and/or new enterprise technologies. We find rampant re-use of the design structure of widely adopted models such as Nolan’s Stage of Growth Model, Crosby’s Grid, and Capability Maturity Model (CMM). We also identify three dominant views of maturity models and provide guidelines for various approaches to constructing maturity models with a standard vocabulary. We finally propose using process theories and configurational approaches to address the main theoretical criticisms with regard to maturity models and conclude with some recommendations for maturity model developers.

Intelligent User Assistance for Automated Data Mining Method Selection

In any data science and analytics project, the task of mapping a domain-specific problem to an adequate set of data mining methods by experts of the field is a crucial step. However, these experts are not always available and data mining novices may be required to perform the task. While there are several research efforts for automated method selection as a means of support, only a few approaches consider the particularities of problems expressed in the natural and domain-specific language of the novice. The study proposes the design of an intelligent assistance system that takes problem descriptions articulated in natural language as an input and offers advice regarding the most suitable class of data mining methods. Following a design science research approach, the paper (i) outlines the problem setting with an exemplary scenario from industrial practice, (ii) derives design requirements, (iii) develops design principles and proposes design features, (iv) develops and implements the IT artifact using several methods such as embeddings, keyword extractions, topic models, and text classifiers, (v) demonstrates and evaluates the implemented prototype based on different classification pipelines, and (vi) discusses the results’ practical and theoretical contributions. The best performing classification pipelines show high accuracies when applied to validation data and are capable of creating a suitable mapping that exceeds the performance of joint novice assessments and simpler means of text mining. The research provides a promising foundation for further enhancements, either as a stand-alone intelligent assistance system or as an add-on to already existing data science and analytics platforms.

The missing G

Artificial general intelligence (AGI) is not a new notion, but it has certainly been gaining traction in recent years, and academic as well as industry resources are redirected to research in AGI. The main reason for this is that current AI techniques are limited as they are designed to operate in specific problem-domains, following meticulous preparation. These systems cannot operate in an unknown environment or under conditions of uncertainty, reuse knowledge gained in another problem domain, or autonomously learn and understand the problem-domain. We shall call AI systems capable of such feats artificial general intelligent (AGI) systems. The three tasks of this paper are to provide a working definition of the term AGI, examine the “missing G”, i.e., the set of abilities that current AI systems lack and whose implementation will result in a basic AGI system, and consider different approaches, including a hybrid one, to a comprehensive solution for an AGI.

A Tool Support for Model-Driven Development: An Industrial Case Study from a Measurement Domain

End-user programming may utilize Domain-Specific Modeling Languages (DSMLs) to develop applications in the form of models, using only abstractions found in a specific problem domain. Indeed, the productivity benefits reported from Model-Driven Development (MDD) are hard to ignore, and a number of MDD solutions are flourishing. However, not all stories from industry on MDD are successful. End-users, without having software development skills, are more likely to introduce software errors than professional programmers. In this study, we propose and encourage other DSML developers to extend the development of DSML with tool support. We believe the programming tools (e.g., debugger, testing tool, refactoring tool) are also needed for end-users to ensure the proper functioning of the products they develop. It is imperative that domain experts are provided with tools that work on the abstraction level that is familiar to them. In this paper, an industrial experience is presented for building various tools for usage in MDD. Debugger, automated testing infrastructure, refactoring, and other tools were implemented for Sequencer, a DSML. Our experience with the implementation of tool support for MDD confirms that these tools are indispensable for end-user programming in practice, and that implementing those tools might not be as costly as expected.

Nitelik Çıkarımı Yöntemlerinin Türkçe Metinlerin Sınıflandırılmasına Etkisi

Feature extraction is the most important preprocessing step of text classification task. Effects of preprocessing techniques on text mining for English have been extensively studied. However, studies for Turkish are limited and generally belong to a specific problem domain. In this study, we investigate the effects of feature extraction techniques on four different Turkish text classification problems including news classification, spam e-mail detection, sentiment analysis, and author detection to show the differences and similarities among the problems. We also propose a new feature selection method to reduce feature space. The experimental analysis has showed that, stopword removal improves classification performance. However, stemming does not make any positive effect on classification accuracy. The most successful term weighting methods are tf and tf*idf. The proposed feature selection method improves classification performance and has higher accuracy than the well-known methods. 

Classification of Isolated Volcano-Seismic Events Based on Inductive Transfer Learning

Domain-specific problems where data collection is an expensive task are often represented by scarce or incomplete data. From a machine learning perspective, this type of problems has been addressed using models trained in different specific domains as the starting point for the final objective-model. The transfer of knowledge between domains, known as transfer learning (TL), helps to speed up training and improve the performance of the models in problems with limited amounts of data. In this letter, we introduce a TL approach to classify isolated volcano-seismic signals at “Volcan de Fuego” , Colima (Mexico). Using the well-known convolutional architecture (LeNet) as a feature extractor and a representative data set containing regional earthquakes, volcano-tectonic earthquakes, long-period events, volcanic tremors, explosions, and collapses, our proposal compares the generalization capabilities of the models when we only fine-tune the upper layers and fine-tune overall of them. Compared with the other state-of-the-art techniques, classification systems based on TL approaches provide good generalization capabilities (attaining nearly 94% of events correctly classified) and decreasing computational time resources.

A High-Order Neutrosophic-Neuro-Gradient Descent Algorithm-Based Expert System for Time Series Forecasting

This article proposed a novel hybrid time series forecasting model using neutrosophic set (NS) theory, artificial neural network (ANN) and gradient descent algorithm. This study deals with three main problems of time series dataset, viz., representation of time series dataset using NS, three degrees of memberships of NS together, and generation of the forecasting results. To resolve these three domain specific problems, this study advocated the application of neutrosophic-neuro-gradient approach. NS theory was utilized to represent the uncertainty associated with time series dataset and was referred to as neutrosophic time series (NTS). In NTS, various decision rules were created in the form of IF-THEN rules, which were termed as neutrosophic entropy decision rules (NEDRs). An ANN-based architecture took NEDRs as input to evolve the forecasting results. To improve the performance of ANN and to obtain optimal forecasting results, this study additionally utilized the gradient descent algorithm to minimize the differences between the calculated and target output values during the simulation. The proposed model was verified and validated with three different datasets, including TAIFEX index, university enrollment dataset of Alabama and Taiwan Stock Exchange Corporation (TSEC) weighted index. Experimental results showed that the proposed model outperformed existing benchmark models with average forecasting error rates of 1.02%, 0.74% and 1.27% for the TAIFEX, university enrollment and TSEC, respectively.

Hinge Epistemology, Radical Skepticism, and Domain Specific Skepticism

This paper explores how hinge epistemology (specifically, Duncan Pritchard’s brand of hinge epistemology) might fruitfully be applied not only to the problem of radical skepticism, but also to certain domain specific (or ‘local’) skepticisms, and in particular, moral skepticism. The paper explains the idea of a domain specific skepticism, and how domain specific skepticisms contrast with radical skepticism. I argue that a domain specific skeptical problem can be resolved in just the same way as radical skepticism, if there are hinge commitments within that domain. I then suggest that there are hinge commitments in the moral domain, and use this to address a moral skeptical problem due to our apparent inability to know moral nihilism to be false.

Convolutional Neural Networks for Recognition of Lymphoblast Cell Images.

This paper presents the recognition for WHO classification of acute lymphoblastic leukaemia (ALL) subtypes. The two ALL subtypes considered are T-lymphoblastic leukaemia (pre-T) and B-lymphoblastic leukaemia (pre-B). They exhibit various characteristics which make it difficult to distinguish between subtypes from their mature cells, lymphocytes. In a common approach, handcrafted features must be well designed for this complex domain-specific problem. With deep learning approach, handcrafted feature engineering can be eliminated because a deep learning method can automate this task through the multilayer architecture of a convolutional neural network (CNN). In this work, we implement a CNN classifier to explore the feasibility of deep learning approach to identify lymphocytes and ALL subtypes, and this approach is benchmarked against a dominant approach of support vector machines (SVMs) applying handcrafted feature engineering. Additionally, two traditional machine learning classifiers, multilayer perceptron (MLP), and random forest are also applied for the comparison. The experiments show that our CNN classifier delivers better performance to identify normal lymphocytes and pre-B cells. This shows a great potential for image classification with no requirement of multiple preprocessing steps from feature engineering.

Sentiment Classification Using Negative and Intensive Sentiment Supplement Information

Traditional methods of annotating the sentiment of an unlabeled document are based on sentiment lexicons or machine learning algorithms, which have shown low computational cost or competitive performance. However, these methods ignore the semantic composition problem displaying in several ways such as negative reversing and intensification. In this paper, we propose a new method for sentiment classification using negative and intensive sentiment supplementary information, so as to exploit the linguistic feature of negative and intensive words in conjunction with the context information. Particularly, our method can solve the domain-specific problem without relying on the external sentiment lexicons. Experimental results on two real-world datasets demonstrate the effectiveness of our proposed method.

A systematic review on deep learning architectures and applications

AbstractThe amount of digital data in the universe is growing at an exponential rate, doubling every 2 years, and changing how we live in the world. The information storage capacity and data requirement crossed the zettabytes. With this level of bombardment of data on machine learning techniques, it becomes very difficult to carry out parallel computations. Deep learning is broadening its scope and gaining more popularity in natural language processing, feature extraction and visualization, and almost in every machine learning trend. The purpose of this study is to provide a brief review of deep learning architectures and their working. Research papers and proceedings of conferences from various authentic resources (Institute of Electrical and Electronics Engineers, Wiley, Nature, and Elsevier) are studied and analyzed. Different architectures and their effectiveness to solve domain specific problems are evaluated. Various limitations and open problems of current architectures are discussed to provide better insights to help researchers and student to resume their research on these issues. One hundred one articles were reviewed for this meta‐analysis of deep learning. From this analysis, it is concluded that advanced deep learning architectures are combinations of few conventional architectures. For example, deep belief network and convolutional neural network are used to build convolutional deep belief network, which has higher capabilities than the parent architectures. These combined architectures are more robust to explore the problem space and thus can be the answer to build a general‐purpose architecture.

Clustering approaches for high‐dimensional databases: A review

Data mining is an inevitable task in most of the emerging computing technologies as it debilitates the complexity of datasets by rendering a better insight. Moreover, it entails the efficacy to envisage ingeniously the vast and heterogeneous datasets and thus delineates substantial knowledge from the abundance of data by pragmatic implementation of suitable algorithm. There are galore of algorithms in literature for this purpose. Furthermore, clustering is widely used techniques to analyze the data within the purview of data mining and thus it became as a motivational impetus for the authors to survey the existing literature on this topic rigorously and have consequently identified various key parameters so that concomitant improvement can be possible while selecting a best fit clustering algorithm pertaining to a specific problem domain. Furthermore, clustering, classification and association rule mining are akin and indispensable to data mining and owing to these authors have also included interrelation and intertwining among these terms so that this work will presage chunk of help for the researchers working in this field. The present study also envisages and manifests the challenges associated with the clustering algorithms for two‐ and high‐dimensional databases in a flamboyant fashion. Over and above, this work identifies key parametric attributes to assess the clustering algorithms which in turn benevolent the existing work and paves the way for profound future research in this realm.This article is categorized under:Technologies > Structure Discovery and ClusteringTechnologies > ClassificationTechnologies > Association RulesFundamental Concepts of Data and Knowledge > Big Data Mining

Great Deluge Based Hyper-heuristics for Solving Real-world University Examination Timetabling Problem: New Data set and Approach

The university examination timetabling problem is one of challenging optimization problems. Its NP-hard nature makes this problem attractive to be studied, especially in the field of operation research and artificial intelligence. In the literature, the state-of-the-art approach for solving examination timetabling problem is meta-heuristics. However, this approach has limitation, i.e. the need for intensive problem-specific parameter tuning. To cope with this problem, a relatively new approach namely hyper-heuristics was proposed. Different from meta-heuristics that search upon solution space, hyper-heuristics search upon low-level space. This strategy makes hyper-heuristics more generic that works over cross-domain, compared to meta-heuristics that usually designed for specific problem domain. This paper reports the success of solving real-world university examination timetabling problem in Institut Teknologi Sepuluh Nopember using hyper-heuristics based on great deluge algorithm. The main contributions of this study are two folds: a new dataset and new approach for solving examination timetabling problem. The computational results show that the proposed algorithm could produce much better solutions compared to the solutions generated manually. In addition, the proposed algorithm also outperforms two benchmarking algorithms, namely hill climbing and simulated annealing algorithms.

Optimized K-Means Clustering Model based on Gap Statistic

Big data has become famous to process, store and manage massive volumes of data. Clustering is an essential phase in big data analysis for many real-life application areas uses clustering methodology for result analysis. The data clustered sets have become a challenging issue in the field of big data analytics. Among all clustering algorithm, the K-means algorithm is the most widely used unsupervised clustering approach as seen from past. The K-means algorithm is the best adapted for deciding similarities between objects based on distance measures with small datasets. Existing clustering algorithms require scalable solutions to manage large datasets. However, for a particular domain-specific problem the initial selection of K is still a significant concern. In this paper, an optimized clustering approach presented which is calculated the optimal number of clusters (k) for specific domain problems. The proposed approach is an optimal solution based on the cluster performance measure analysis based on gab statistic. By observation, the experimental results prove that the proposed model can efficiently enhance the speed of the clustering process and accuracy by reducing the computational complexity of the standard k-means algorithm which achieves 76.3%.