Ing. Radek Mařík, CSc.

Automated journalism is experiencing rapid advancements, prompting concerns among certain journalists who fear potential job displacement by robotic reporters. Conversely, there is a cohort of journalists embracing this progress, viewing the involvement of robotic reporters as a time-efficient measure and an avenue to engage in more creative forms of journalism. This article aims to explore topics in Czech Republic news that lend themselves to automation based on perplexity. Currently, the robotic reporter can generate articles in few domains (e.g., stock market news, fuel prices news, and traffic accident reports), while further exploration is underway for other areas. However, generating e.g. cultural news through automation remains challenging due to its less recurrent nature, as evidenced by perplexity measurements reflecting text complexity. Perplexity method can be used to determine which journalistic area is suitable for automated text. In this article, we focus on the Czech environment and the possibilities of automation and the reaction of selected Czech journalists who encounter automated texts in newsrooms. While automated journalism demonstrates superiority over human journalists in certain aspects, it still falls short in terms of creativity, an aspect where artificial intelligence remains unable to replace humans. Nevertheless, the robotic reporter can serve as an invaluable assistant to editors and reporters in their work. Consequently, the newsroom landscape is evolving, accommodating the integration of automated content editors.

Active learning is a powerful technique that accelerates model learning by iteratively expanding training data based on the model’s feedback. This approach has proven particularly relevant in natural language processing and other machine learning domains. While active learning has been extensively studied for conventional classification tasks, its application to more specialized tasks like neural coreference resolution has the potential for improvement. In our research, we present a significant advancement by applying active learning to the neural coreference problem, and setting a benchmark of 39% reduction in required annotations for training data. Simultaneously, it preserves performance compared to the original model trained on the full data. We compare various uncertainty sampling techniques along with Bayesian modifications of coreference resolution models, conducting a comprehensive analysis of annotation efforts. The results demonstrate that the best-performing techniques seek to maximize label annotation in previously chosen documents, showcasing their effectiveness and preserving performance.

Supervised learning of classifiers for text classification and sentiment analysis relies on the availability of labels that may be either difficult or expensive to obtain. A standard procedure is to add labels to the training dataset sequentially by querying an annotator until the model reaches a satisfactory performance. Active learning is a process that optimizes unlabeled data records selection for which the knowledge of the label would bring the highest discriminability of the dataset. Batch active learning is a generalization of a single instance active learning by selecting a batch of documents for labeling. This task is much more demanding because plenty of different factors come into consideration (i. e. batch size, batch evaluation, etc.). In this paper, we provide a large scale study by decomposing the existing algorithms into building blocks and systematically comparing meaningful combinations of these blocks with a subsequent evaluation on different text datasets. While each block is known (warm start weights initialization, Dropout MC, entropy sampling, etc.), many of their combinations like Bayesian strategies with agglomerative clustering are first proposed in our paper with excellent performance. Particularly, our extension of the warm start method to batch active learning is among the top performing strategies on all datasets. We studied the effect of this proposal comparing the outcomes of varying distinct factors of an active learning algorithm. Some of these factors include initialization of the algorithm, uncertainty representation, acquisition function, and batch selection strategy. Further, various combinations of these are tested on selected NLP problems with documents encoded using RoBERTa embeddings. Datasets cover context integrity (Gibberish Wackerow), fake news detection (Kaggle Fake News Detection), categorization of short texts by emotional context (Twitter Sentiment140), and sentiment classification (Amazon Reviews). Ultimately, we show that each of the active learning factors has advantages for certain datasets or experimental settings.

The present study aims to outline new, more adjusted approaches of research addressing social complexity of past societies. In doing so, we use varied evidence to detect major 'leap events' in the history of ancient Egypt which were reflected by the state administration and its fluctuating complexity. The archaeological and inscriptional evidence shows that crucial changes in history had a non-linear, punctuated character. To reveal their true character, newly developed mathematical models have been applied. The analyses of early complex civilisations have made a noticeable progress recently. The current scholarship pays significant attention to a processual approach, description of the dynamics and its interpretation against the specific background formed by varied datasets originating from disciplines such as archaeology, history, art history, philology or environmental sciences to name but a few of the most relevant ones. Within this context, Old Kingdom Egypt evidence is reassessed using specific methods of analysis and interpretation. The ancient Egyptian Old Kingdom (2592-2120 BC), one of the earliest territorial states on this planet, is still frequently considered to be a homogenous continuum of isolated historical events manifested in various forms of architecture, art or religion. Some recent studies applied to its study put emphasis on a non-linear, 'punctuated approach' which appears to provide some new important perspectives on this traditional problem. The application of modern mathematical methods based on Hidden Markov Models and Social Network Analysis significantly changes this view. These methods have the potential to detail a vivid, heterogenous process of historical progress as a punctuated equilibria model, as a non-linear system with changing dynamics of its development in time. In this process, human agency, the rise and fluctuation of complexity and particular strategies of different social groups played significant roles and can be detected with the help of impartial approaches. The emerging picture can be used not only to describe the evolution of a past society but also for comparative purposes when studying the dynamics of past or present societies.

Supervised classification of texts relies on the availability of reliable class labels for the training data. However, the process of collecting data labels can be complex and costly. A standard procedure is to add labels sequentially by querying an annotator until reaching satisfactory performance. Active learning is a process of selecting unlabeled data records for which the knowledge of the label would bring the highest discriminability of the dataset. In this paper, we provide a comparative study of various active learning strategies for different embeddings of the text on various datasets. We focus on Bayesian active learning methods that are used due to their ability to represent the uncertainty of the classification procedure. We compare three types of uncertainty representation: i) SGLD, ii) Dropout, and iii) deep ensembles. The latter two methods in cold- and warm-start versions. The texts were embedded using Fast Text, LASER, and RoBERTa encoding techniques. The methods are tested on two types of datasets, text categorization (Kaggle News Category and Twitter Sentiment140 dataset) and fake news detection (Kaggle Fake News and Fake News Detection datasets). We show that the conventional dropout Monte Carlo approach provides good results for the majority of the tasks. The ensemble methods provide more accurate representation of uncertainty that allows to keep the pace of learning of a complicated problem for the growing number of requests, outperforming the dropout in the long run. However, for the majority of the datasets the active strategy using Dropout MC and Deep Ensembles achieved almost perfect performance even for a very low number of requests. The best results were obtained for the most recent embeddings RoBERTa

A large number dimensions may cause a variety of problems in real-world applications: some dimensions might be redundant and can worsen the quality of the workflow output, and, in the vast majority of exercises with datasets, data are distributed along a highly nonlinear manifold whose structure is unknown. This paper focuses on analyzing the outputs of nonlinear dimensionality reduction, or Manifold Learning, techniques. We introduce three meaningful measures that are capable of providing context behind projections onto lower-dimensional spaces. The measures will enable us to compare techniques with each other and assist in choosing suitable hyperparameters. Moreover, we propose to view projections from the standpoint of simplicial complex distortion. In connection to that, we establish the process of a dimension-agnostic graph-based data tessellation technique that builds a simplicial skeleton of high-dimensional data. Alongside our new tessellation technique, we evaluate the proposed quality measures on the Delaunay-tessellation-based simplicial approximations of manifolds.

Hidden Markov models (HMMs) are one of the most widely used unsupervised-learning algorithms for modeling discrete sequential data. Traditionally, most of the applications of HMMs have utilized only models of order 1 because higher-order models are computationally hard to train. We reformulate HMMs using tensor decomposition to efficiently build higher-order models with the use of stochastic gradient descent. Based on this, we propose a new modified version of a training algorithm for HMMs, especially suitable for high-order HMMs. Further, we show its capabilities and convergence on synthetic data.

Kernel density estimation is a complex task that plays an essential role in a variety of applications. In this paper, we introduce an approach to the task that converts the problem of bandwidth evaluation in the Parzen-window-like framework into the non-parametric evaluation of a fine-grained density estimate which can then be scaled by means of the Scale-Space theory to achieve the desired level of smoothness. The detailed estimate is realized through the Delaunay space tessellation method and properties of its output simplices. Additionally, in the experimental part of the paper, we showcase the new method and demonstrate its outputs at various scales, reaching results that perceivably outperform its counterparts.

The sophisticated administration in the age of the pyramid builders offers a remarkable time span for research and a unique opportunity to analyse the dynamics of a complex society in a diachronic perspective. Contrary to traditional approaches relying on statistics and logic, we will present an overview of our achievements in society development reconstruction covering structural aspects using complex network analysis (CNA). Our research is based on diachronic occurrences of titles held by officials of various social status. The presence of a social stratification structure can be demonstrated by specialized bipartite (people – titles) network visualizations. Results aimed at reconstructing the structure of Old Kingdom top-level society and at the detection of powerful dignitaries using CNA and particularly community detection methods are also discussed. The approaches are exemplified by selected case studies on influential persons from the Fifth and Sixth dynasties.

Egyptology is becoming a strategic scientific discipline in that it is instructive for contemporary civilization and makes it possible to avoid critical mistakes in its development. However, classical Egyptology tools need to be supplemented by modern procedures that are more efficient in terms of the speed of the production of results and the processing of much more extensive data volumes, delivering more accurate and reliable evaluation of the findings. Since the time of Plato, cybernetics has been providing methods by which models based on observations of the environment or sensor data are created to reflect the properties of systems and environments, and actions that change the environment are generated. Many of these practices, which include elements of learning and deduction techniques, are developed within the field of artificial intelligence. Methods that specialize in analyzing relationships that can be modeled by graph theory are now widely developed within the framework of complex network analysis. To emphasize the increasing rate of application of these scientific methods to cover new challenges in processing ancient Egyptian data, the concept of cyber-Egyptology has been introduced recently. Cyber-Egyptology deals with the interpretation of ancient Egyptian data and Egyptologists' observations using techniques based on the principles of cybernetics, artificial intelligence, and complex network analysis that might be modified to cope with a limited amount of complex structures containing uncertainties and missing entries. We offer an overview of a selected set of such methods and their application to Old Kingdom data.

The state of a state machine is a path between two actions; however, it is the rest of the world who walks the path. The development of a web application, especially of an information system, starts with use cases, i.e., model scenarios of how users interact with the application and its entities. The goal of this paper is to turn the use cases into a useful specification and automatically convert them into a model layer of a web application, in our case using finite automata. BPMN (Business Process Model and Notation) provides a graphical syntax to capture the use cases, which is based on the theoretical background of Petri Nets. However, because BPMN does not capture the state of the modeled entities, it is impractical to use it as a specification of a persistent storage and model layers of the web application. To overcome this problem, we propose a new STS algorithm to infer a finite automaton that implements a chosen participant in a BPMN diagram that represents a given entity of the web application.

Time-varying requirements of users on communication push mobile operators to increase density of base stations. However, the dense deployment of conventional static base stations (SBSs) is not always economical, for example, when periods of peak load are short and infrequent. In such cases, several Fying base stations (FlyBSs) mounted on unmanned aerial vehicles can be seen as a convenient substitution for the dense deployment of SBSs. This paper focuses on maximization of user satisfaction with provided data rates. To this end, we propose an algorithm that associates users with the most suitable SBS/FlyBS and finds optimal positions of all FlyBSs. Furthermore, we investigate the performance of two proposed approaches for the joint association and positioning based on the genetic algorithm (GA) and particle swarm optimization (PSO). It is shown that both solutions improve the satisfaction of users with provided data rates in comparison with a competitive approach. We also demonstrate trade-offs between the GA and the PSO. While the PSO is of lower complexity than the GA, the GA requires a slightly lower number of active FlyBSs to serve the users.

We improve a method of multitree-like graph visualization using a spanning tree-driven layout technique with constraints specified by layers and the ordering of groups of nodes within layers. We propose a new method of how the order of subtrees selected by the driving spanning tree can be derived from the actual edge crossings. Such a subtree order leads to additional decreasing of total edge crossings from 1% to 50%. This depends on the shape of the processed graph, ranging from a pure tree to a general acyclic graph. Our achievements are demonstrated using several datasets containing up to millions of people, species, or services. The proposed subtree ordering method of layered graphs that are similar to acyclic multitrees retains the generating of acceptable layouts in almost linear time.

Community detection in bipartite networks with overlapping communities carries difficult problems when it comes to complex network analysis. In this paper, we propose a new model for generating such graphs. We combine several approaches based on stochastic block models using edge probabilities following the Poisson distribution. The proposed model can be reduced into their original model versions, such as a model for a bipartite graph with non-overlapping communities only. We present results of the generator. Its performance is evaluated using several known community detection techniques. The evaluation criterion assesses both a community’s identification and their overlaps.

State machines and a relational database may look like completely unrelated tools, yet they form an interesting couple. By supporting them with well-established architectural patterns and principles, we built a model layer of a web application which utilizes the formal aspects of the state machines to aid the development of the application while standing on traditional technologies. The layered approach fits well with existing frameworks and the Command-Query Separation pattern provides a horizontal separation and compatibility with various conceptually distinct storages, while the overall architecture respects RESTful principles and the features of the underlying SQL database. The integration of the explicitly specified state machines as first-class citizens provides a reliable connection between the well-separated formal model and the implementation; it enables us to use visual comprehensible formal models in a practical and effective way, and it opens new possibilities of using formal methods in application development and business process modeling.

This article discusses the decomposition of an object space described by a large number of binary and ordinal features. It is assumed that no class labels are known and assigned to objects. The proposed method is a combination of clustering and community detection methods and allows for the construction of overlapping clusters. Subsets of relevant features are detected using redundant informa- tion. Feature dependency measures are based on information theory tools. Achieved results are demonstrated on data sets characterizing officials in the Old Kingdom of ancient Egypt.

We propose a new method of node positioning for huge layered graphs specified by layers and fixed ordering of nodes within layers. We assume that the assignments of nodes to the layers and the order of nodes within the layers are provided by other suitable methods capable of processing multitree like networks. The node positioning method is based on the force-driven approach with barrier-like repulsive forces that avoids the quadratic complexity of traditional methods. We demonstrate achievements on several datasets containing up to millions of people or species. The proposed layout method of layered graphs that are close to acyclic multitrees creates aesthetically acceptable layouts in linear time.

A detection of anomaly events in a non-stationary time series resembling vibration aspects is often based on the signal envelope. Both commonly used approaches, i.e. the Hilbert transform and the empirical mode decomposition, lead to high processing power requirements. As anomalies are assumed to be rare, we propose simplified methods based on extreme value theory that enable trend corrections and anomaly event occurrence detection. Both methods are based on the upper and lower envelope approximations that are constructed using maxima and minima limits of the signal.

We leverage a method of multitree graph visualization using a spanning tree-driven layout technique with constraints specified by layers and the ordering of groups of nodes within layers. We propose a new method of how the driving spanning tree can be selected using graph blocks and the simplex method. We demonstrate achievements using several datasets containing up to 10 6 people or species. The proposed layout method of layered graphs that are similar to acyclic multitrees creates aesthetically acceptable layouts in linear time.

Almost every practical technique dealing with a detection of events in signals uses a technique of thresholding. The quality of the method performance often depends heavily on the threshold selection. In this paper, assuming that the signal values are bounded, we propose a simple practical method of threshold determination based on signal maxima sampling. We also present the relation between the ad hoc version of the method and the extreme value theory. In fact, it is shown that the method follows the extreme value theory conclusions if their asymptotic assumptions are satisfied by the given practical problem. The experimental results confirm the performance of the method we have applied in projects to different signal domains for many years. © 2018, Springer International Publishing AG.

Automated techniques of anomaly and outlier detection usually rely on thresholding. The quality of the method performance often depends heavily on the threshold selection. In this paper, assuming that the signal values are bounded, we propose a simple, practical method of threshold determination based on signal maxima sampling. It can be shown that only modeling block maxima is a wasteful approach to extreme value determination if other data on extremes are available. In our contribution, we reformulate the method of threshold selection based on blocks into a method using so-called threshold models. Threshold models rely on extracting the peak values from signal values which exceed a certain threshold. The experimental results confirm the performance of the method.

Deployment of commercial computer networks sets high requirements for procedures, tools, and approaches for comprehensive testing of these networks. However, in spite of the great efforts of many researchers, the process of test design/generation still tends to be unstructured and bound to the personal experience and/or intuition of individual engineers. To address the problem, this paper introduces an approach of automated generation of abstract test cases based on the concept of multilayer networks. Test cases of that kind cover network infrastructures, including individual components and component-to-component interactions on all coexisting architectural layers, and provide information for subsequent analysis to ensure that the used formal model is consistent with respect to test requirements.

The paper proposes a brand-new approach to processing, explanation, and interpretation of ancient Egyptian raw data set covering the period of the Old Kingdom (2700-2180 BC), taking into account the identification and analysis of social processes. At that time the gradual transformation of Egyptian society from a kingdom to a state took place. Methods of complex networks, referenced also as social network analysis, have been used for quantification of a number of historical aspects, there are only scattered publications dealing with late history of Egypt. Regrettably, nobody has applied SNA to the Old Kingdom context yet.

While a visual unconstrained tree structure planar layout design is easy to implement, a visualization of a tree with constraints on node ranks and their ordering within ranks leads to a difficult combinatorial problem. A genealogical graph, such as family tree, can be taken as an example of such a case. Classical ancestor trees, descendant trees, Hourglass charts, and their visual variants such as node-link diagrams or fan charts are suitable for assessment of peoples relationships when one is focused on a particular person and his/her direct ancestors and descendants. Such tree-based representations miss a broader context of relationships and do not allow the quick assessment of several interlinked families together.We propose a new undirected tree-driven layout technique for layered multitree graph visualizations producing constraints on node layers and ordering of groups of nodes within layers. The computed constraints can be mapped, at least partially, into the DOT language property directives used by the Graphviz toolbox. We demonstrate achievements on several datasets containing up to 39000 people.

By focusing on the development of the administration and its changes and turning points during the Old Kingdom period, this contribution proposes a dynamic model of bureaucratic evolution at that time punctuated by several ground-breaking changes which can be viewed as adaptations to the social dynamics. Endeavouring to find a way and method of visualizing this major observation, we focus on the stratified vizieral titulary, since it was directly related to the society’s development and reflected all major shifts and modifications. The Hidden Markov Model method has been adjusted in a way that enables its robust application to a limited volume of input data. As a consequence, its results make it possible to quantify the development of the administration structure in ancient Egypt. The numerical results are interpreted and compared to the state-of-art knowledge in Egyptology. In particular, several landmarks matching the perspective of the ‘punctuated equilibrium’ theory are discussed.

The watershed transformation is widely used, a powerful tool for the segmentation of 2D and 3D images. In this paper we propose a fast linear one sweep algorithm generating all possible extreme weighted connected components in 1D real value signal.We also show how such a post-order list of connected components can be converted into a component tree in linear time. Furthermore, we demonstrate how some features of the connected components can be computed again in linear time.

Pattern and anomaly detection in time series or sequences is an active field of research. Most of the work in the field is dedicated to multivariate sequences of real numbers or they only focus on pattern recognition in simple sequences. This work introduces hidden Markov models and applies new methods which are suitable for the processing of binary or categorical multivariate sequences in which individual elements are considered to be expressions of the same system. Subsequently, proposed methods are applied in the analysis of the dataset of viziers and their titles from the fourth to the sixth dynasties of ancient Egypt.

By focusing on methods of systematic data collection and the complex network analysis, the paper proposes a new research approach to processing and interpretation of "raw data" from the Old Kingdom period by using semi-automatic methods.

One of the major goals of multilayer networks is providing proper and suitable representations of complex systems with many interdependent components, which, in turn, might interact through many different channels. However, the terminology referring to systems with multiple different relations has not yet reached a consensus. This work introduces a simple taxonomy of multilayer networks. Four different dimensions characterize the basic metrics of structural properties. Based on the taxonomy, an appropriate formal definition is presented.

Formal methods based on abstract models are becoming more and more important in the domain of complex computer networks. On the other hand, processes of design documentation transformation into the formal models are still bound to the skills and ingenuity of individual engineers. Moreover, the human factor involved in data transformation represents a major bottleneck due to the tendency of computer networks to be more and more complex. To address this problem, this work introduces a formal model based on the concept of multilayer networks for applying a system methodology to network analysis and an appropriate presentation format of architecture descriptions as a possible part of detailed design documentation that could allow automated generation of trusted formal multilayer models based on this documentation.

We pose a problem with the aim to design a complete set of data consistency checking rules. The context domain is assumed to be a set of entities characterized by property values and a set of relations among entities. The source data, called observations, are assumed to be fused into a network of entities, further referred to as the network model. Using the bounded model checker Alloy, we focus on a practical design of a set of all consistency rules that are sufficient to ensure that the result network model is consistent.

Classical ancestor trees, descendant trees, Hourglass charts, and their visual variants such as node-link diagrams or fan charts are suitable for assessment of people’s relationships when one is focused on a particular person (the so-called main person) and his/her direct ancestors and descendants. Such tree-based representations miss a broader context of relationships and do not allow quick assessment of several interlinked families together. We propose utilization of directed acyclic graph visualizations with constraints specified by layers and ordering of groups of nodes within layers. The computed constraints can be mapped, at least partially, into the DOT language property directives used by the Graphviz toolbox. We demonstrate achievements on datasets containing 1600 people (a private family tree collection) and 3000 people (an Egyptology database of officials from 4th, 5th, and 6th dynasty).

This article presents the processing of radiographic data acquired using the position-sensitive hybrid semiconductor pixel detector Timepix. Measurements were made on thin samples at the medical ion-synchrotron HIT [1] in Heidelberg (Germany) with a 221 MeV proton beam. The charge is energy by the particles crossing the sample is registered for generation of image contrast. Experimental data from the detector were processed for derivation of the energy loss of each proton using calibration matrices. The interaction point of the protons on the detector were determined with subpixel resolution by model fitting of the individual signals in the pixeled matrix. Three methods were used for calculation of these coordinates: Hough transformation, 2D Gaussian fitting and estimate the 2D mean. Parameters of calculation accuracy and calculation time are compared for each method. The final image was created by method with best parameters.

While a visual unconstrained tree structure planar layout design is easy to implement, a visualization of a tree with constraints on node ranks and their ordering within ranks leads to a difficult combinatorial problem. A genealogical graph can be taken as an example of such a case. We propose a new method of tree-driven graph node layout.

Deployment of network/distributed systems sets high requirements for procedures, tools and approaches for the complex testing of these systems. This work provides a survey of testing activities with regard to these systems based on standards and actual practices for both software-based and distribution (network) aspects. On the basis of this survey, we determine formal testing procedures/processes which cover these aspects, but which are not contrary to both aspects. The next step, based on the analysis of the implementation phase of System Development Life Cycle, determines a formal model for these processes.

Formal methods based on abstract models are becoming more and more important in the domain of complex computer networks. On the other hand, processes of design documentation transformation into the formal models are still bound to the skills and ingenuity of individual engineers. Moreover, the human factor involved in data transformation represents a major bottleneck due to the tendency of computer networks to be more and more complex. To address this problem, this work introduces a possible appropriate presentation format of architecture descriptions as a part of detailed design documentation that could allow automated development of trusted formal models for analysis and verifying of complex computer networks.

We cannot make a clear statement to which extent we have approached the state, which in terms of evaluation of the history of mankind is already dangerously close to a period when a process known as collapse of complex society is launched. The contribution provides an overview of dynamic models and evaluation of nepotism.

Viber is a proprietary communication protocol growing in popularity, on mobile devices especially. It represents a trend of replacing traditional protocols with proprietary messaging applications, employing cloud infrastructure support. It also constitutes a challenge to the networking community due to its unknown properties and opaque and highly distributed nature. We present observations-based analysis of the protocol function, topology, communication patterns and vulnerabilities, derived from traffic captures by simple data mining.

Deployment of distributed systems sets high requirements for procedures and tools for the complex testing of these systems. This work introduces a formal four - layered model for test generation mission o n the basis of the component based approach and the concept of layered networks . Based on this model, we de scribe the test generation stratégy that covers every interaction from the end user requirements on all co existing architectural layers , and checks the internal consistency of the system technical specifications with respect to the end user requirements. The next step introduce s the Prolog based approach to representing this model and the requirements coverage stratégy.

When building complex applications the only way not to get lost is to split the application into simpler components. Current programming languages, including object oriented ones, offer very good utilities to create such components. However, when the components are created, they need to be connected together. Unluckily, these languages are not a very suitable tool for that. To help with composition of the components we introduce cascade – a dynamic acyclic structure built from blocks, inspired by the Function Block approach. The cascade generates itself on-the-fly during its evaluation to match requirements specified by input data and automatically orders an execution of the individual blocks. Thus the structure of a given cascade does not need to be predefined entirely during its composing/implementation and fixed during its execution as it is usually assumed by the most approaches. It also provides a real-time and fully automatic visualization of all blocks and their connections to ease debugging and an inspection of the application.

Smalldb uses a non-deterministic parametric finite automaton combined with Kripke structures to describe lifetime of an entity, usually stored in a traditional SQL database. It allows to formally prove some interesting properties of resulting application, like access control of users, and provides primary source of metadata for various parts of the application, for example automatically generated user interface and documentation.

A method for structure discovery is described. We assume that an unknown composite structure is sampled and observed as a set of small relational structure fractions. Each fraction is described by a set of atoms and their relations. We propose a technique that allows fusing these fractions into large components or the structure itself. The method is demonstrated using an example taken from the genealogy domain. Our approach of the solution is based on SAT. We show that the composition can be effectively performed using the Alloy tool, which is a bounded model checker. The search through the possible solutions is guided by the unique identities of observations.

Web applications are developed and modified often so fast that architects, developers, and managers lose their control over quality of such software products. Reverse engineering techniques focused on different aspects of software implementation might help in keeping comprehension to implementation of web applications at appropriate level required for fulfilling change request. We focus our effort on a reconstruction of implicit associations among GUI data items and back-end database entities. Our approach is based on an association analysis and mining using a synchronized log of GUI events and related counterparts of SQL statements. Recovered dependecies between GUI and back-end databased can be utilized advantageously in an automated design of web application data flow testing.

We present a method designed to solve the problem of automatic color shade grading for industrial inspection of randomly textured multicolored ceramic tiles. Our grading method is based on the comparison of color histograms and we present results of correctly grading several series of ceramic tiles, the differences of which were at the threshold of human perception.

The ceramic tile manufacturing process is now completely automated, with the exception of the final stage of production concerned with visual inspection. In this paper, the authors describe an integrated system developed for the detection of defects on color ceramic tiles and for the color grading of defect-free tiles. The results suggest that the performance is adequate for providing a basis for a viable commercial visual inspection system. (C) 1998 Elsevier Science S.A, All rights reserved.

We present a method designed to solve the problem of automatic color grading for industrial inspection of plain and patterned ceramic tiles, We discuss problems me mere confronted with, like the temporal and spatial variation of the illumination, and the ways we dealt with them. Then, we present results of correctly grading a series of ceramic tiles, the differences of which were at the threshold of human perception.

The detection of abnormalities in stochastic textures is a challenging task in comparison to locating faults in replicated texture images. The authors demonstrate a technique based in the spatial and feature domains and detect abnormal formulations across the surface of randomly-textured images. The technique is based on colour clustering followed by structural processing with morphological operators.