Modelling and Development of Intelligent Systems

Progressive developments in the world of Information and Communications Technology open up many frontiers in the educational sector. One of such is adaptive e-learning systems, which is currently attracting a lot of research and development. Several conceptualisations and implementations rely on single parameters, or at most three or four parameters. This is not sufficient to account for the wide range of factors which can affect the learning process in an unconventional learning environment such as the web. Being able to choose relevant parameters for personalisation in different learning scenarios is vital to accommodate a wide range of these factors. In this paper, we’ll do a review of the basic concepts and components of an adaptive e-learning system. Afterwards, we’ll present a model of an adaptive e-learning system which generates a specialised curriculum for a learner based on a multi-parameter approach, thereby allowing for more choices in the process of creating a personalised and learner-oriented experience for such user. This will involve assembling (and/or suggesting) learning resources encompassed in a general curriculum and adapting it to specific personalities and preferences of users. The degree of adaptation (of the curriculum) is dependent on a weighted algorithm matching user features (relevant in each learning scenario) to the corresponding features of available learning resources.

To successfully achieve the goal of providing global access to quality education, the Information and Communications Technology (ICT) sector has provided tremendous advances in virtual/online learning. One of such advances is the availability of digital learning resources. However, to successfully accommodate learner peculiarities and predispositions, traditional online learning is gradually being transformed from a one-size-fits-all paradigm towards personalised learning. This transformation requires that learning resources are treated not as static content, but dynamic entities, which are reusable, portable across different platforms, and ultimately adaptive to user needs. This article takes a review of how digital learning resources are modelled in adaptive hypermedia systems to achieve adaptive learning, and we highlight prospects of future work.

The product line domain and agile software development created a well-received product line concept with complex toolchains and the agile development idea that fosters a rather pragmatic development style with less documentation. Both parts have been included in two student projects targeting the development of a configurator product line. This paper presents requirements for configurator product lines and evaluates the two projects towards these requirements. The positive results of the projects lead to the proposal to broaden the support and acceptance of special project types in the product line domain where the source code and structure is enough to understand and maintain the project. We even propose this to be the better solution.

Conceptual models are the foundation for many modern intelligent systems, as well as a theoretical basis for conducting more in-depth scientific research. Various information sources (e.g., databases, spreadsheets data, and text documents, etc.) and the reverse engineering procedure can be used for creation of such models. In this paper, we propose an approach to support the conceptual model engineering based on the analysis and transformation of tabular data from CSV files. Industrial safety inspection (ISI) reports are used as examples for spreadsheets data analysis and transformation. The automated conceptual model engineering involves five steps and employs the following software: TabbyXL for extraction of canonical (relational) tables from arbitrary spreadsheet data in the CSV format; Personal Knowledge Base Designer (PKBD) for generation of conceptual model fragments based on analysis and transformation of canonical tables, and aggregating these fragments into domain model. Verification of the approach was carried out on the corpus containing 216 spreadsheets extracted from six ISI reports. The obtained conceptual models can be used in the design of knowledge bases.

Assessing the nonlinearity of one signal, system, or dependence of one signal on another is of great importance in the design process. The article proposes an algorithm for simplified nonlinearity estimation of digital signals. The solution provides detailed information to constructors about existing nonlinearities, which in many cases is sufficient to make the correct choice of processing algorithms. The programming code of the algorithm is presented and its implementation is demonstrated on a set of basic functions. Several steps to further development of the proposed approach are outlined.

Common people are increasingly acquiring health information depending on general search engines which are still far from being effective in dealing with complex consumer health queries. One prime and effective method in addressing this problem is using Learning to Rank (L2R) techniques. In this paper, an investigation on aggregation over field-based L2R models is made. Rather than combining all potential features into one list to train a L2R model, we propose to train a set of L2R models each using features extracted from only one field and then apply aggregation methods to combine the results obtained from each model. Extensive experimental comparisons with the state-of-the-art baselines on the considered data collections confirmed the effectiveness of our proposed approach.

The major challenge in e-learning is the assessment as a tool to measure students’ knowledge. In this regard an intelligent system for generation and evaluation of e-learning tests using integer programming is proposed. The described system aims to determine a number of questions with different degree of difficulty from a predefined set of questions that will compose the test. It allows also generating tests with different level of complexity. To realize the selection of the questions for different levels of tests two optimization models are proposed. Both of these models are of linear integer programming. The first of them determines the minimum number of questions by selecting among more difficult questions, while the second one aims to maximize the number of questions by selecting among less difficult questions. The proposed intelligent system for generating and evaluating e-learning tests with different levels of complexity is implemented as web-based application. The numerical testing of the developed prototype of the intelligent system for generation of tests for e-learning purposes is demonstrated in a web programming course.

The General Data Protection Regulation (GDPR) has entered into force in the European Union (EU) since 25 May 2018 in order to satisfy present difficulties related to private information protection. This regulation involves significant structural for companies, but also stricter requirements for personal data collection, management, and protection. In this context, companies need to create smart solutions to allow them to comply with the GDPR and build a feeling of confidence in order to map all their personal data. In these conditions, cognitive computing could be able to assist companies extract, protect and anonymize sensitive structured and unstructured data. Therefore, this article proposes a framework that can serve as an approach or guidance for companies that use cognitive computing methods to meet GDPR requirements. The goal of this work is to examine the smart system as a data processing and data protection solution to contribute to GDPR compliance.

One of the major global concerns nowadays is definitely the pollution. The effects are more and more visible as time passes, our daily activities affecting the environment more than they should. Pollution has effects on air, water and soil. According to the European Economic Area (EEA), air pollution is the main cause of premature death in 41 European nations. Their studies found high levels of air pollutants in Poland that came second on the list, topped by Turkey. Therefore, in this article we aim to determine a model for greenhouse gas (GHG) emissions and atmospheric pollutants in Poland based on a set of data retrieved from a European statistics website.

This paper presents an automatic facial analysis system which is able to perform gender detection, hair segmentation and geometry detection, color attributes extraction (hair, skin, eyebrows, eyes and lips), accessories (eyeglasses) analysis from facial images. For the more complex tasks (gender detection, hair segmentation, eyeglasses detection) we used state of the art convolutional neural networks, and for the other tasks we used classical image processing algorithms based on geometry and appearance models. When data was available, the proposed system was evaluated on public datasets. An acceptance study was also performed to assess the performance on the system in real life scenarios.

Industrial furnaces consume a large amount of energy and their operating points have a major influence on the quality of the final product. Designing a tool that analyzes the combustion process, fluid mechanics and heat transfer and assists the work done during energy audits is then of the most importance.

This work proposes a hybrid model for such a tool, having as its base two white-box models, namely a detailed Computational Fluid Dynamics (CFD) model and a simplified Reduced-Order (RO) model, and a black-box model developed using Machine Learning (ML) techniques.

The preliminary results presented in the paper show that this composite model is able to improve the accuracy of the RO model without having the high computational load of the CFD model.

Software testing is a very expensive and critical activity in the software systems’ life-cycle. Finding software faults or bugs is also time-consuming, requiring good planning and a lot of resources. Therefore, predicting software faults is an important step in the testing process to significantly increase efficiency of time, effort and cost usage.

In this study we investigate the problem of Software Faults Prediction (SFP) based on Neural Network. The main contribution is to empirically establish the combination of Chidamber and Kemer software metrics that offer the best accuracy for faults prediction with numeric estimations by using feature selection. We also proposed a conceptual framework that integrates the model for fault prediction.

Handwritten digit recognition is an important subarea in the object recognition research area. Support vector machines represent a very successful recent binary classifier. Basic support vector machines have to be improved in order to deal with real-world problems. The introduction of soft margin for outliers and misclassified samples as well as kernel function for non linearly separably data leads to the hard optimization problem of selecting parameters for these two modifications. Grid search which is often used is rather inefficient. In this paper we propose the use of one of the latest swarm intelligence algorithms, the fireworks algorithm, for the support vector machine parameters tuning. We tested our approach on standard MNIST base of handwritten images and with selected set of simple features we obtained better results compared to other approaches from literature.