Maintaining data warehouses and ETL processes is becoming increasingly difficult. For this reason, we introduce a similarity measure on ETL processes, based on the edit distance of a graph, which models the process. We show both the exact way how to calculate it and heuristic approaches to compute the estimated similarity more quickly. We propose methods to improve graph edit distance based on the assumption that the ETL process model is a directed acyclic graph.

In this paper we present a cryptographic protocol for a seller - buyer problem and in particular we prove that non-emptiness of a semi-commutative set defined by mappings involved in the protocol is a decidable problem.

We consider a version of the k-path vertex cover problem that asks for the minimum weight subset C of vertices of a graph G such that every path on k vertices in G has at least one vertex in common with C. We present two dynamic algorithms solving this problem on interval graphs. The first one works on general interval graphs but is in practice limited to small values of k.
The second algorithm computes minimum weight vertex cover for arbitrary k on proper interval graph G = (V,E) in time O(|V|^2|E|).

There are many open questions in this area of computer science that are very important from the perspective of the social media marketing. Among them is: "how to write the messages that are 'popular and liked'"? In this paper we will model and investigate one possible aspect of this issue: does sentiment of the social media post correlates with social engagement of fan base? We have modeled sentiment scoring of social media post using lexicon - based method and by state of the art convolutional neural network. The evaluation of those models has been performed using social media Twitter accounts of five worldknown politicians and celebrities, four brands, two bloggers and two users. We have investigated the various statistical dependencies between sentiment - based scores and engagement scores values. Basing on results we can concluded that number of favorites or shares (both are among the most popular engagement scoring methods that are present in most social media platforms) is not dependent on the sentiment of the message. It does not matter if posts have positive or negative sentiment. The results we have obtained are very important especially for researchers and business entities who utilizes social media platform. Large number of social media scoring algorithms utilizes some kind of binary sentiment analysis associated with social engagement scoring. Our results are strong indicators that two popular sentiment analysis methods should not be used as the predictors of mentioned social engagement scores. Our research can be easily reproduced because we publish both our data and source code of programs we used for evaluation.

In this note we present theory which facilitates the use of Conley index algorithms for cubical multivalued maps constructed from maximal dimensional cubes in the setting when cubes of arbitrary dimension are permitted.

Independent Component Analysis (ICA) is a method for searching the linear transformation that minimizes the statistical dependence between its components. Most popular ICA methods use kurtosis as a metric of independence (non-Gaussianity) to maximize, such as FastICA and JADE. However, their assumption of fourth-order moment (kurtosis) may not always be satisfied in practice. One of the possible solution is to use third-order moment (skewness)  instead of kurtosis, which was applied in ICA_SG and EcoICA. In this paper we present a competitive approach to ICA based on the Split Generalized Gaussian distribution (SGGD), which is well adapted to heavy-tailed as well as asymmetric data. Consequently, we obtain a method which works better than the classical approaches, in both cases: heavy tails and non-symmetric data.

One of the most important optimization tasks in the industry at the current time is the object packing problem. Although several methods have been developed for the purpose of solving it, they are usually only able to optimize placement locally and as such are heavily dependent on the choice of the initial setting -- hence the need for trying out multiple possible starting points, which impacts algorithm running time. In this paper we present a neural network-based model which provides sensible starting points in a linear time.

We investigate performance of a gradient descent optimization (GR) applied to the traffic signal setting problem and compare it to genetic algorithms. We used neural networks as metamodels evaluating quality of signal settings and discovered that both optimization methods produce similar results, e.g., in both cases the accuracy of neural networks close to local optima depends on an activation function (e.g., TANH activation makes optimization process converge to different minima than ReLU activation).

Regularizing the gradient norm of the output of a neural network is a powerful technique, rediscovered several times. This paper presents evidence that gradient regularization can consistently improve classification accuracy on vision tasks, using modern deep neural networks, especially when the amount of training data is small. We introduce our regularizers as members of a broader class of Jacobian-based regularizers. We demonstrate empirically on real and synthetic data that the learning process leads to gradients controlled beyond the training points, and results in solutions that generalize well.

In this paper, we propose a simple, fast and easy to implement algorithm LOSSGRAD (locally optimal step-size in gradient descent), which automatically modifies the step-size in gradient descent during neural networks training. Given a function f, a point x, and the gradient ▽_xf of f, we aim to find the step-size h which is (locally) optimal, i.e. satisfies:

h = arg min f(x - t▽_xf).
            t≥0

Making use of quadratic approximation, we show that the algorithm satisfies the above assumption. We experimentally show that our method is insensitive to the choice of initial learning rate while achieving results comparable to other methods.

We apply the optimal transport distance to construct two goodness of fit tests for (univariate) normality. The derived statistics are then compared with those used by the Shapiro-Wilk, the Anderson-Darling and the Cramer-von Mises tests. In particular, we preform Monte Carlo experiments, involving computations of the test power against some selected alternatives and wide range of sample sizes, which show efficiency of the obtained test procedures.

In this paper we discuss a class of  AutoEncoder based generative models based on one dimensional sliced approach. The idea is based on the reduction of the discrimination between samples to one-dimensional case.

Our experiments show that methods can be divided into two groups. First consists of methods which are a modification of standard normality tests, while the second is based on classical distances between samples.

It turns out that both groups are correct generative models, but the second one gives a slightly faster decrease rate of Frechet Inception Distance (FID).

We present a novel modification of context encoder loss function, which results in more accurate and plausible inpainting. For this purpose, we introduce gradient attention loss component of loss function, to suppress the common problem of inconsistency in shapes and edges between the inpainted region and its context. To this end, the mean absolute error is computed not only for the input and output images, but also for their derivatives. Therefore, model concentrates on areas with larger gradient, which are crucial for accurate reconstruction. The positive effects on inpainting results are observed both for fully-connected and fully-convolutional models tested on MNIST and CelebA datasets.

The article introduces a new set of Polish word embeddings, built using KGR10 corpus, which contains more than 4 billion words. These embeddings are evaluated in the problem of recognition of temporal expressions (timexes) for the Polish language. We described the process of KGR10 corpus creation and a new approach to the recognition problem using Bidirectional Long-Short Term Memory (BiLSTM) network with additional CRF layer, where specific embeddings are essential. We presented experiments and conclusions drawn from them.

Infinite length of trajectories is an almost universal assumption in the theoretical foundations of reinforcement learning. In practice learning occurs on finite trajectories. In this paper we examine a specific result of this disparity, namely a strong bias of the time-bounded Every-visit Monte Carlo value estimator. This manifests as a vastly different learning dynamic for algorithms that use value predictors, including encouraging or discouraging exploration. We investigate these claims theoretically for a one dimensional random walk, and empirically on a number of simple environments. We use GAE as an algorithm involving a value predictor and evolution strategies as a reference point.

In this paper we present a method with closed analytic formula of stitching aligned images.

It is obtained by choosing a statistically optimal global color change of one part of image.  This approach, due to its numerical efficiency, is especially well-suited for merging big amount of satellite images into a single map.

Moreover, we present solution of a general problem, how to find an optimal shift by v of data Y with respect to v from V, so that the dataset  X, Y+v is maximally statistically consistent. We show that the solution is given in a closed analytic form.

Natural language inference (NLI) is a central problem in natural language processing (NLP) of predicting the logical relationship between a pair of sentences. Lexical knowledge, which represents relations between words, is often important for solving NLI problems. This knowledge can be accessed by using an external knowledge base (KB), but this is limited to when such a resource is accessible. Instead of using a KB, we propose a simple architectural change for attention based models. We show that by adding a skip connection from the input to the attention layer we can utilize better the lexical knowledge already present in the pretrained word embeddings. Finally, we demonstrate that our strategy allows to use an external source of knowledge in a straightforward manner by incorporating a second word embedding space in the model.

Neural networks are prone to adversarial attacks. In general, such attacks deteriorate the quality of the input by either slightly modifying most of its pixels, or by occluding it with a patch. In this paper, we propose a method that keeps the image unchanged and only adds an adversarial framing on the border of the image. We show empirically that our method is able to successfully attack state-of-the-art methods on both image and video classification problems. Notably, the proposed method results in a universal attack which is very fast at test time.

Credibility is an important part of any quality scoring method. Scoring quality of a piece of information in regard to other messages gives an additional and crucial point of view. This quality dimension is especially useful when evaluating information produced by sensors. Often grouped in a network, a message from one sensor can be evaluated using other sources in this network which can highlight problematic messages leading to improved decision making or optimising maintenance operations.

This paper considers the problem of defining credibility on sensor data: including definition of confirmation and invalidation for a piece of information in the case of the more challenging event-type messages. The proposed method aims to define a network of correlated sensors from which the relevant messages are extracted and then aggregated.

Different propositions for the final aggregation step of confirming and invalidating messages leads to the definition of a flexible framework that can be adapted to different scenarios.

An example is presented based on a real dataset that includes sensors from railway domain.

This paper is dedicated to employ novel technique of deep learning for machines failures prediction. General idea of how to transform sensor data into suitable data set for prediction is presented. Then, neural network architecture that is very successful in solving such problems is derived. Finally, we present a case study for real industrial data of a gas turbine, including results of the experiments.

The use of machine learning methods in the case of incomplete data is an important task in many scientific fields, like medicine, biology, or face recognition. Typically, missing values are substituted with artificial values that are estimated from the known samples, and the classical machine learning algorithms are applied. Although this methodology is very common, it produces less informative data, because artificially generated values are treated in the same way as the known ones. In this paper, we consider a probabilistic representation of missing data, where each vector is identified with a Gaussian probability density function, modeling the uncertainty of absent attributes. This representation allows to construct an analogue of RBF kernel for incomplete data. We show that such a kernel can be successfully used in regression SVM. Experimental results confirm that our approach capture relevant information that is not captured by traditional imputation methods.

A novel model of dynamically programmed attributed regular grammars, DPAR, for the ECG diagnosis justification purposes is presented in the paper. A formal model, power properties and a case of DPAR grammar are described. The formalism of DPAR grammars allows to differentiate between certain subclasses of ECG phenomena.

In this paper we present a theorem concerning an equivalent statement of the Jacobian Conjecture in terms of Picard-Vessiot extensions. Our theorem completes the earlier work of T. Crespo and Z. Hajto which suggested an effective criterion for detecting polynomial automorphisms of affine spaces. We show a simplified criterion and give a bound on the number of wronskians determinants which we need to consider in order to check if a given polynomial mapping with non-zero constant Jacobian determinant is a polynomial automorphism. Our method is specially efficient with cubic homogeneous mappings introduced and studied in fundamental papers by H. Bass, E. Connell, D.Wright and L. Drużkowski

For a positive integer n, n-expandable deep pushdown automata always contain no more than n occurrences of non-input symbols in their pushdowns during any computation. As its main result, the present paper demonstrates that these automata are as powerful as the same automata with only two non-input pushdown symbols - $ and #, where # always appears solely as the pushdown bottom. The paper demonstrates an infinite hierarchy of language families that follows from this main result. The paper also points out that if # is the only non-input symbol in these automata, then they characterize the family of regular languages. In its conclusion, the paper suggests open problems and topics for the future investigation.

Size of the data is often a challenge in real-life applications. Especially when working with time series data, when next sample is produced every few milliseconds and can include measurement from hundreds of sensors, one has to take the dimensionality of the data into consideration. In this work, we compare various dimensionality reduction methods for time series data and check their performance on failure detection task. We work on sensory data coming from existing machines.

We discuss a new framework for solving extreme classification (i.e., learning problems with an extremely large label space), in which we reduce the original problem to a structured prediction problem. Thanks to this we can obtain learning algorithms that work under a strict time and space budget. We mainly focus on a recently introduced algorithm, referred to as LTLS, which is to our best knowledge the first truly logarithmic time and space (in the number of labels) method for extreme classification. We compare this algorithm with two other approaches that also rely on transformation to structured prediction problems. The first algorithm encodes original labels as binary sequences. The second algorithm follows the label tree approach. The comparison shows the trade-off between computational complexity (in time and space) and predictive performance.

There is a strong research eort towards developing models that can achieve state-of-the-art results without sacrificing interpretability and simplicity. One of such is recently proposed Recursive Random Support Vector Machine (R2SVM) model, which is composed of stacked linear models. R2SVM was reported to learn deep representations outperforming many strong classiffiers like Deep Convolutional Neural Network. In this paper we try to analyze it both from theoretical and empirical perspective and show its important limitations.Analysis of similar model Deep Representation Extreme Learning Machine (DrELM) is also included. It is concluded that models in its current form achieves lower accuracy scores than Support Vector Machine with Radial Basis Function kernel.

Deep neural networks are currently among the most commonly used classifiers. Despite easily achieving very good performance, one of the best selling points of these models is their modular design – one can conveniently adapt their architecture to specific needs, change connectivity patterns, attach specialised layers, experiment with a large amount of activation functions, normalisation schemes and many others. While one can find impressively wide spread of various configurations of almost every aspect of the deep nets, one element is, in authors’ opinion, underrepresented – while solving classification problems, vast majority of papers and applications simply use log loss. In this paper we try to investigate how particular choices of loss functions affect deep models and their learning dynamics, as well as resulting classifiers robustness to various effects. We perform experiments on classical datasets, as well as provide some additional, theoretical insights into the problem. In particular we show that L1 and L2 losses are, quite surprisingly, justified classification objectives for deep nets, by providing probabilistic interpretation in terms of expected misclassification. We also introduce two losses which are not typically used as deep nets objectives and show that they are viable alternatives to the existing ones.

Object ranking is one of the most relevant problems in the realm of preference learning and ranking. It is mostly tackled by means of two different techniques, often referred to as pairwise and pointwise ranking. In this paper, we present a case study in which we systematically compare two representatives of these techniques, a method based on the reduction of ranking to binary classification and so-called expected rank regression (ERR). Our experiments are meant to complement existing studies in this field, especially previous evaluations of ERR. And indeed, our results are not fully in agreement with previous findings and partly support different conclusions.

In this work the subject of the application of clustering as a knowledge extraction method from real-world data is discussed. The authors analyze an influence of different clustering parameters on the quality of the created structure of rules clusters and the efficiency of the knowledge mining process for rules / rules clusters. The goal of the experiments was to measure the impact of clustering parameters on the efficiency of the knowledge mining process in rulebased knowledge bases denoted by the size of the created clusters or the size of the representatives. Some parameters guarantee to produce shorter/longer representatives of the created rules clusters as well as smaller/greater clusters sizes.

Continuous vector representations, as a distributed representations for words have gained a lot of attention in Natural Language Processing (NLP) field. Although they are considered as valuable methods to model both semantic and syntactic features, they still may be improved. For instance, the open issue seems to be to develop different strategies to introduce the knowledge about the morphology of words. It is a core point in case of either dense languages where many rare words appear and texts which have numerous metaphors or similies. In this paper, we extend a recent approach to represent word information. The underlying idea of our technique is to present a word in form of a bag of syllable and letter n-grams. More specifically, we provide a vector representation for each extracted syllable-based and letter-based n-gram, and perform concatenation. Moreover, in contrast to the previous method, we accept n-grams of varied length n. Further various experiments, like tasks-word similarity ranking or sentiment analysis report our method is competitive with respect to other state-of-theart techniques and takes a step toward more informative word representation construction.

Robust mixture models approaches, which use non-normal distributions have recently been upgraded to accommodate data with fixed bounds. In this article we propose a new method based on uniform distributions and Cross-Entropy Clustering (CEC). We combine a simple density model with a clustering method which allows to treat groups separately and estimate parameters in each cluster individually. Consequently, we introduce an effective clustering algorithm which deals with non-normal data.

Recent methods for learning word embeddings, like GloVe or Word2-Vec, succeeded in spatial representation of semantic and syntactic relations. We extend GloVe by introducing separate vectors for base form and grammatical form of a word, using morphosyntactic dictionary for this. This allows vectors to capture properties of words better. We also present model results for word analogy test and introduce a new test based on WordNet.

Due to rapid growth of computational power and demand for faster and more optimal solution in today's manufacturing, machine learning has lately caught a lot of attention. Thanks to it's ability to adapt to changing conditions in dynamic environments it is perfect choice for processes where rules cannot be explicitly given. In this paper proposes on-line supervised learning approach for optimal scheduling in manufacturing. Although supervised learning is generally not recommended for dynamic problems we try to defeat this conviction and prove it's viable option for this class of problems. Implemented in multi-agent system algorithm is tested against multi-stage, multi-product flow-shop problem. More specifically we start from dening considered problem. Next we move to presentation of proposed solution. Later on we show results from conducted experiments and compare our approach to centralized reinforcement learning to measure algorithm performance.

Mutation testing is considered as one of the most effective quality improvement technique by assessing the strength of the actual test suite. If no test is able to kill a given mutant, this means that the tests are not strong enough and we need to write additional one that will be able to kill this mutant. However, mutation testing is very time consuming. In this paper we investigate if it is possible to reduce the scope of the mutation analysis by running it only on the new or changed part of the code. Using data from the real open-source projects we analyze if there is a relation between mutation scope reduction and effectiveness of the mutation analysis.

Support Vector Machines (SVM) with RBF kernel is one of the most successful models in machine learning based compounds biological activity prediction. Unfortunately, existing datasets are highly skewed and hard to analyze. During our research we try to answer the question how deep is activity concept modeled by SVM. We perform analysis using a model which embeds compounds’ representations in a low-dimensional real space using near neighbour search with Jaccard similarity. As a result we show that concepts learned by SVM is not much more complex than slightly richer nearest neighbours search. As an additional result, we propose a classification technique, based on Locally Sensitive ashing approximating the Jaccard similarity through minhashing technique, which performs well on 80 tested datasets (consisting of 10 proteins with 8 different representations) while in the same time allows fast classification and efficient online training.

This paper presents a novel global thresholding algorithm for the binarization of documents and gray-scale images using Cross-Entropy Clustering. In the first step, a gray-level histogram is constructed, and the Gaussian densities are fitted. The thresholds are then determined as the cross-points of the Gaussian densities. This approach automatically detects the number of components (the upper limit of Gaussian densities is required).

This work presents the step by step tutorial for how to use cross entropy clustering for the iris segmentation. We present the detailed construction of a suitable Gaussian model which best fits for in the case of iris images, and this is the novelty of the proposal approach. The obtained results are promising, both pupil and iris are extracted properly and all the information necessary for human identification and verification can be extracted from the found parts of the iris.

Deep learning is a field of research attracting nowadays much attention, mainly because deep architectures help in obtaining outstanding results on many vision, speech and natural language processing – related tasks. To make deep learning effective, very often an unsupervised pretraining phase is applied. In this article, we present experimental study evaluating usefulness of such approach, testing on several benchmarks and different percentages of labeled data, how Contrastive Divergence (CD), one of the most popular pretraining methods, influences network generalization.

This paper concerns classification of high-dimensional yet small sample size biomedical data and feature selection aimed at reducing dimensionality of the microarray data. The research presents a comparison of pairwise combinations of six classification strategies, including decision trees, logistic model trees, Bayes network, Na¨ıve Bayes, k-nearest neighbours and sequential minimal optimization algorithm for training support vector machines, as well as seven attribute selection methods: Correlation-based Feature Selection, chi-squared, information gain, gain ratio, symmetrical uncertainty, ReliefF and SVM-RFE (Support Vector Machine-Recursive Feature Elimination). In this paper, SVMRFE feature selection technique combined with SMO classifier has demonstrated its potential ability to accurately and efficiently classify both binary and multiclass high-dimensional sets of tumour specimens.

In this article we present the use of sparse representation of a signal and incoherent dictionary learning method for the purpose of network traffic analysis. In learning process we use 1D INK-SVD algorithm to detect proper dictionary structure. Anomaly detection is realized by parameter estimation of the analyzed signal and its comparative analysis to network traffic profiles. Efficiency of our method is examined with the use of extended set of test traces from real network traffic. Received experimental results confirm effectiveness of the presented method.

Balancing exploratory and exploitative behavior is an essential dilemma faced by adaptive agents. The challenge of finding a good trade-off between exploration (learn new things) and exploitation (act optimally based on what is already known) has been largely studied for decision-making problems where the agent must learn a policy of actions. In this paper we propose the engaged climber method, designed for solving the exploration-exploitation dilemma. The solution consists in explicitly creating two different policies (for exploring or for exploiting), and to determine the good moments to shift from the one to the other by the use of notions like engagement and curiosity.

The selection of data representation and metric for a given data set is one of the most crucial problems in machine learning since it affects the results of classification and clustering methods. In this paper we investigate how to combine a various data representations and metrics into a single function which better reflects the relationships between data set elements than a single representation-metric pair. Our approach relies on optimizing a linear combination of selected distance measures with use of least square approximation. The application of our method for classification and clustering of chemical compounds seems to increase the accuracy of these methods.

This paper presents a formalised description of the models of influence propagation in social networks introduced in the classic paper of Kempe et al. The formal framework that we propose clarifies the structure of the most popular propagation models and helps rigorously re-establish the essential results concerning the problem of influence maximisation. We also introduce new models of propagation and show how they fit into the general picture. In particular, we focus on models that capture either positive or negative effects of resisting influence on individual’s future resistance.

We propose a novel model of multilinear filtering based on a hierarchical structure of covariance matrices – each matrix being extracted from the input tensor in accordance to a specific set-theoretic model of data generalization, such as derivation of expectation values. The experimental analysis results presented in this paper confirm that the investigated approaches to tensor-based data representation and processing outperform the standard collaborative filtering approach in the ‘cold-start’ personalized recommendation scenario (of very sparse input data). Furthermore, it has been shown that the proposed method is superior to standard tensor-based frameworks such as N-way Random Indexing (NRI) and Higher-Order Singular Value Decomposition (HOSVD) in terms of both the AUROC measure and computation time.

In the paper a contour ensemble image segmentation concept is presented. It bases on the previously observed relationship between contours and classifiers. Because of the specificity of the active contour segmentation the method requires a special procedure to obtain ensemble members with desired properties. In this work it is achieved by early stopping of randomized optimization algorithm. The results of the method are illustrated with a practical problem of heart ventricle segmentation by means of active potential contours. Automatically found contours may be of use in a process of pulmonary embolism diagnosis.

Multithreshold Entropy Linear Classifier (MELC) is a recent classifier idea which employs information theoretic concept in order to create a multithreshold maximum margin model. In this paper we analyze its consistency over multithreshold linear models and show that its objective function upper bounds the amount of misclassified points in a similar manner like hinge loss does in support vector machines. For further confirmation we also conduct some numerical experiments on five datasets.

We present a new subspace clustering method called SuMC (Subspace Memory Clustering), which allows to efficiently divide a dataset D  RN into k 2 N pairwise disjoint clusters of possibly different dimensions. Since our approach is based on the memory compression, we do not need to explicitly specify dimensions of groups: in fact we only need to specify the mean number of scalars which is used to describe a data-point. In the case of one cluster our method reduces to a classical Karhunen-Loeve (PCA) transform. We test our method on some typical data from UCI repository and on data coming from real-life experiments.

We consider the Boussinesq PDE perturbed by a time-dependent forcing. Even though there is no smoothing effect for arbitrary smooth initial data, we are able to apply the method of self-consistent bounds to deduce the existence of smooth classical periodic solutions in the vicinity of 0. The proof is non-perturbative and relies on construction of periodic isolating segments in the Galerkin projections.

In this paper, a novel probabilistic tracking method is proposed. It combines two competing models: (i) a discriminative one for background classification; and (ii) a generative one as a track model. The model competition, along with a combinatorial data association, shows good signal and background noise separation. Furthermore, a stochastic and derivative-free method is used for parameter optimization by means of the Covariance Matrix Adaptation Evolutionary Strategy (CMA-ES). Finally, the applicability and performance of the particle trajectories reconstruction are shown. The algorithm is developed for NA61/SHINE data reconstruction purpose and therefore the method was tested on simulation data of the NA61/SHINE experiment.

Transistor level simulation of the CPU, while very accurate, brings also the performance challenge. MOS6502 CPU simulation algorithm is analysed with several optimisation techniques proposed. Application of these techniques improved the transistor level simulation speed by a factor of 3–4, bringing it to the levels on par with fastest RTL-level simulations so far.

Deep understanding of microprocessor architecture, its internal structure and mechanics of its work is essential for engineers in the fields like computer science, integrated circuit design or embedded systems (including microcontrollers). Usually the CPU architecture is presented at the level of ISA, functional decomposition of the chip and data flows. In this paper we propose more tangible, interactive and effective approach to present the CPU microarchitecture. Based on the recent advancements in simulation of MOS6502, one of the most successful microprocessor of all times, that started the personal computing revolution, we present the CPU visualisation framework. The framework supports showing CPU internals at various levels (from single transistor, through logic gates, ending with registers, operation decoders and ALU). It allows for execution of real code and detailed analysis of fetch–decode–execute cycle, measurement of cycles per operation or measurement of the CPU activity factor. The analysis means provided by this framework will also enable us to propose the transistor level simulation speed improvements to the model in the future.

This paper presents an optimal scheduling solution for a case of agents sharing a resource. The amount of resource can not satisfy all agents at once and in case of runout there is a penalty. Each agent randomly toggle its state between requiring and not requiring the resource. Using the knowledge of previous state and probability of change, the scheduling algorithm is able to calculate optimal number of concuring agents for one turn, that minimizes possibility of collision yet provides as much throughput as possible. Several different scheduling strategies are tested. The optimal solution adapts automatically to the value of probability of change. Further experiments show that optimality is retained if only the average probability of a set of agents is known. A case of practical application is provided.

In this paper, we introduce a new kind of automata systems, called state-synchronized automata systems of degree n. In general, they consists of n pushdown automata, referred to as their components. These systems can perform a computation step provided that the concatenation of the current states of all their components belongs to a prescribed control language. As its main result, the paper demonstrates that these systems characterize the family of recursively enumerable languages. In fact, this characterization is demostrated in both deterministic and nondeterministic versions of these systems. Restricting their components, these systems provides less computational power.

A subset S of vertices of a graph G = (V,E) is called a k-path vertex cover if every path on k vertices in G contains at least one vertex from S. Denote by Ψ_k (G) the minimum cardinality of a k-path vertex cover in G and form a sequence Ψ (G) = (Ψ₁(G), Ψ₂ (G), . . . , Ψ_|V|(G)), called the path sequence of G. In this paper we prove necessary and sufficient conditions for two integers to appear on fixed positions in Ψ(G). A complete list of all possible path sequences (with multiplicities) for small connected graphs is also given.

The paper concerns the construction scheme of Direct Digital Synthesis (DDS) generator based on widely developed Field Programmable Gate Arrays (FPGA) technology. Firstly, the division of all generators is presented regarding the fact whether they generate sinusoidal or non-sinusoidal signals and whether the they have positive or negative feedback. Next chapter concerns how to generate frequency directly and indirectly synthesis and how to integrate these types of sytheses into the hybrid one. According to the blueprint, DDS generator if fully digital, which facilitates the control enabling to change almost at will the oscillation while the system works. Next chapters contain the conclusions and some proposals regarding practical application. DDS synthesizer has been made in accordance to the assumptions of the project, which was confirmed in practice. Control program enables us to fuse some measuring systems together into one, which facilitates simultaneous gathering of the data from several devices. Hence, the overall efficiency rises and the costs of carried out experiments are reduced. In the case of foreigners, it will be supplied by editors.

We present some typical algorithms used for finding global minimum/ maximum of a function defined on a compact finite dimensional set, discuss commonly observed procedures for assessing and comparing the algorithms’ performance and quote theoretical results on convergence of a broad class of stochastic algorithms.

For the contemporary business, the crucial factor is making smart decisions on the basis of the knowledge hidden in stored data. Unfortunately,m traditional simple methods of data analysis are not sufficient for efficient management of modern enterprizes, because they are not appropriate for the huge and growing amount of the data stored by them. Additionally data usually comes continuously in the form of so-called data stream. The great disadvantage of traditional classification methods is that they assume that statistical properties of the discovered concept are being unchanged, while in real situation, we could observe so-called concept drift, which could be caused by changes in the probabilities of classes or/and conditional probability distributions of classes. The potential for considering new training data is an important feature of machine learning methods used in security applications (spam filtering or intrusion detection) or decision support systems for marketing departments, which need to follow the changing client behavior. Unfortunately, the occurrence of concept drift dramatically decreases classification accuracy. This work presents the comprehensive study on the ensemble classifier approach applied to the problem of drifted data streams. Especially it reports the research on modifications of previously developed Weighted Aging Classifier Ensemble (WAE) algorithm, which is able to construct a valuable classifier ensemble for classification of incremental drifted stream data. We generalize WAE method and propose the general framework for this approach. Such framework can prune an classifier ensemble before or after assigning weights to individual classifiers. Additionally, we propose new classifier pruning criteria, weight calculation methods, and aging operators. We also propose rejuvenating operator, which is able to soften the aging effect, which could be useful, especially in the case if quite ”old” classifiers are high quality models, i.e., their presence increases ensemble accuracy, what could be found, e.g., in the case of recurring concept drift. The chosen characteristics of the proposed frameworks were evaluated on the basis of the wide range of computer experiments carried out on the two benchmark data streams. Obtained results confirmed the usability of proposed method to the data stream classification with the presence of incremental concept drift.

We review three mathematical developments linked with Hilberg’s conjecture – a hypothesis about the power-law growth of entropy of texts in natural language, which sets up a challenge for machine learning. First, considerations concerning maximal repetition indicate that universal codes such as the Lempel-Ziv code may fail to efficiently compress sources that satisfy Hilberg’s conjecture. Second, Hilberg’s conjecture implies the empirically observed power-law growth of vocabulary in texts. Third, Hilberg’s conjecture can be explained by a hypothesis that texts describe consistently an infinite random object.

Consider the problem of approximating a Markov chain by another Markov chain with a smaller state space that is obtained by partitioning the original state space. An information-theoretic cost function is proposed that is based on the relative entropy rate between the original Markov chain and a Markov chain defined by the partition. The state space aggregation problem can be sub-optimally solved by using the information bottleneck method.

Multithreshold Entropy Linear Classifier (MELC) is a density based model which searches for a linear projection maximizing the Cauchy-Schwarz Divergence of dataset kernel density estimation. Despite its good empirical results, one of its drawbacks is the optimization speed. In this paper we analyze how one can speed it up through solving an approximate problem. We analyze two methods, both similar to the approximate solutions of the Kernel Density Estimation querying and provide adaptive schemes for selecting a crucial parameters based on user-specified acceptable error. Furthermore we show how one can exploit well known conjugate gradients and L-BFGS optimizers despite the fact that the original optimization problem should be solved on the sphere. All above methods and modifications are tested on 10 real life datasets from UCI repository to confirm their practical usability.

This paper discusses path controlled grammars—context-free gram-
mars with a root-to-leaf path in their derivation trees restricted by a control
language. First, it investigates the impact of erasing rules on the generative
power of path controlled grammars. Then, it establishes two Chomsky-like
normal forms for path controlled grammars—the first allows unit rules, the
second allows just one erasing rule.

The paper summarizes properties of topological and sequence en- tropy of the Morse shift X_M generated by the Thue-Morse sequence t_M. The first part is an estimation of growth rate of possible subwords in t_M. We show a polynomial upper bound on the number of finite subwords occuring in t_M which is Cn^2log3 for some constant C > 0. In the second part we prove that the sequence entropy of X_M is achieved for the sequence  r(i) = 2²ⁱ − 1.

Various kinds of decipherability of codes, weaker than unique de- cipherability, have been studied since mid-1980s. We consider decipherability of directed figure codes, where directed figures are defined as labelled polyomi- noes with designated start and end points, equipped with catenation operation that may use a merging function to resolve possible conflicts. This setting ex- tends decipherability questions from words to 2D structures. In the present paper we develop a (variant of) domino graph that will allow us to decide some of the decipherability kinds by searching the graph for specific paths. Thus the main result characterizes directed figure decipherability by graph properties.

In a Cantor metric space B^Z, we present a one-sided cellular
automaton which positively answers the question
Does it exist a transitive cellular automaton (B^Z, F) with non-empty set of
strictly temporally periodic points?
The question can be found in a current and recognized literature of the subject.

This paper presents new grammar systems that describe transfor- mations of syntactic structures. They represent two approaches: synchronous grammars and transducers. The systems consist of well-known models such as context-free grammars and finite automata. Particular attention is paid to synchronization of regulated grammars. The paper recalls formal definitions of the systems and discusses theoretical results regarding their generative and accepting power. The last part briefly introduces application perspectives in natural language translation, illustrated by examples of Czech-English translation.

The quadratic loss penalty is a well known technique for optimization and control problems to treat constraints. In the present paper they are applied to handle control bounds in a boundary control problems with semilinear elliptic state equations. Unlike in the case of finite dimensional optimization for infinite dimensional problems the order of convergence could only be roughly estimated, but numerical experiments revealed a clearly better convergence behavior with constants independent of the dimension of the used discretization. The main result in the present paper is the proof of sharp convergence bounds for both, the finite und infinite dimensional problem with a mesh-independence in case of the discretization. Further, to achieve an efficient realization of penalty methods the principle of control reduction is applied, i.e. the control variable is represented by the adjoint state variable by means of
some nonlinear function. The resulting optimality system this way depends only on the state and adjoint state. This system is discretized by conforming linear finite elements. Numerical experiments show exactly the theoretically predicted behavior of the studied penalty technique.

We consider a nonlinear Neumann elliptic equation driven by the p-Laplacian and a Carathéodory perturbation. The energy functional of the problem need not be coercive. Using variational methods we prove an existence theorem and a multiplicity theorem, producing two nontrivial smooth solutions. Our formulation incorporates strongly resonant equations..

In this paper we study the nonlinear elliptic problem with p(x)-Laplacian (hemivariational inequality). We prove the existence of a nontrivial solution. Our approach is based on critical point theory for locally Lipschitz functionals due to ChangIn this paper we study the nonlinear elliptic problem with p(x)-Laplacian (hemivariational inequality). We prove the existence of a nontrivial solution. Our approach is based on critical point theory for locally Lipschitz functionals due to Chang

IWe derive C2 −characterizations for convex, strictly convex, as well as strongly convex functions on full dimensional convex sets. In the cases of convex and strongly convex functions this weakens the well-known openness assumption on the convex sets. We also show that, in a certain sense, the full dimensionality assumption cannot be weakened further. In the case of strictly convex functions we weaken the well-known sufficient C2 −condition for strict convexity to a characterization. Several examples illustrate the results.

In J. J. Ye and D. L. Zhu proposed a new reformulation of a bilevel programming problem which compounds the value function and KKT approaches. In partial calmness condition was also adapted to this new reformulation and optimality conditions using partial calmness were introduced. In this paper we investigate above all local equivalence of the combined reformulation and the initial problem and how constraint qualifications and optimality conditions could be defined for this reformulation without using partial calmness.
Since the optimal value function is in general nondifferentiable and KKT constraints have MPEC-structure, the combined reformulation is a nonsmooth MPEC. This special structure allows us to adapt some constraint qualifications and necessary optimality conditions from MPEC theory using disjunctive form of the combined reformulation. An example shows, that some of the proposed constraint qualifications can be fulfilled.
 

The paper deals with the numerical treatment of the optimal control of drying of materials which may lead to cracks. The drying process is controlled by temperature, velocity and humidity of the surrounding air. The state equations dene the humidity and temperature distribution within a simplied wood specimen for given controls. The elasticity equation describes the internal stresses under humidity and temperature changes. To avoid cracks these internal stresses have to be limited. The related constraints are treated by smoothed exact barrier-penalty techniques. The objective functional of the optimal control problem is of tracking type. Further it contains a quadratic regularization by an energy term for the control variables (surrounding air) and barrier-penalty terms.
The necessary optimality conditions of the auxiliary problem form a coupled system of nonlinear equations in appropriate function spaces. This optimality system is given by the state equations and the related adjoint equations, but also by an approximate projection onto the admissible set of controls by means of barrier-penalty terms. This system is discretized by nite elements and treated iteratively for given controls. The optimal control itself is performed
by quasi-Newton techniques.

The quasidifferential calculus developed by V.F. Demyanov and A.M. Rubinov provides a complete analogon to the classical calculus of differentiation for a wide class of nonsmooth functions. Although this looks at the first glance as a generalized subgradient calculus for pairs of subdifferentials it turns out that, after a more detailed analysis, the quasidifferential calculus is a kind of Fréchet-differentiation whose gradients are elements of a suitable Minkowski–Rådström–Hörmander space. One aim of the paper is to point out this fact. The main results in this direction are Theorem 1 and Theorem 5. Since the elements of the Minkowski–Rådström–Hörmander space are not uniquely determined, we focus our attention in the second part of the paper to smallest possible representations of quasidifferentials, i.e. to minimal representations. Here the main results are two necessary minimality criteria, which are stated in Theorem 9 and Theorem 11.
 

In this paper basic mathematical tasks of coordinate measurement are briefly described and a modied optimization algorithm is proposed. Coordinate measurement devices generate huge data set and require adapted methods to solve related mathematical problems in real time. The proposed algorithm possesses a simplied step size rule and nds the solution of the minimum circumscribed ball fitting after only a nite number The iteration is of the steepest descent type applied to the related distance function. But, in contrast to standard algorithms it uses a modied step size rule that takes into account the specic properties of the occurring objective function. This small dierence in the code improves the performance of the algorithm and it enables real time use of the proposed method in coordinate measurement machines. The eciency of the prosed algorithm will be illustrated by some typical examples.
 

We present the main concept and results of the p-regularity theory (also known as p-factor analysis of nonlinear mappings) applied to nonlinear optimization problems. The approach is based on the construction of p-factor operator. The main result of this theory gives a detailed description of the structure of the zero set of irregular nonlinear mappings. Applications include a new numerical method for solving nonlinear optimization problems and p-order necessary and sufficient optimality conditions. We substantiate the rate of convergence of p-factor method.
 

We consider a Dirichlet boundary value problem driven by the p-Laplacian with the right hand side being a Carathéodory function. The existence of solutions is obtained by the use of a special form of the three critical points theorem.

Automatic differentiation is an often superior alternative to numerical differentiation that is yet unregarded for calculating derivatives in the optimization of imaging optical systems. We show that it is between 8% and 34% faster than numerical differentiation with central difference when optimizing various optical systems.

We present a Lohner type algorithm for the computation of rigorous bounds for the solutions of ordinary differential equations and its derivatives with respect to the initial conditions up to an arbitrary order.

Character segmentation (i.e., splitting the images of handwritten words into pieces corresponding to single letters) is one of the required steps in numerous off-line cursive handwritten word recognition solutions. It is also a very important step, because improperly extracted characters are usually impossible to recognize correctly with currently used methods. The most common method of character segmentation is initial oversegmentation – finding some set of potential splitting points in the graphical representation of the word and then attempting to eliminate the improper ones. This paper contains a list of popular approaches for generating potential splitting points and methods of verifying their correctness.
 

The aim of this study was to describe some parametric estimation methods for the Weibull, gamma and Gompertz distributions and to identify among them estimators the most efficient in practical applications. Techniques which are considered as traditional methods, like the maximum likelihood (MLE) and the method of moments (MM) estimation but also some newer and less commonly used techniques like the Lmoment estimator (LME), least-square estimator (LSE), generalized spacing estimator (GSE) and percentile estimator (PE) were presented. The application of each method was demonstrated in a simulation study using data sets generated for different distribution parameters and sample sizes. Discussed estimators were compared in terms of their efficiency and bias measured by mean-square errors (MSE) based on the simulations results.

In this paper a new theorem about components of the mean squared error of Hierarchical Estimator is presented. Hierarchical Estimator is a machine learning meta-algorithm that attempts to build, in an incremental and hierarchical manner, a tree of relatively simple function estimators and combine their results to achieve better accuracy than any of the individual ones. The components of the error of a node of such a tree are: weighted mean of the error of the estimator in a node and the errors of children, a non-positive term that descreases below 0 if children responses on any example dier and a term representing relative quality of an internal weighting function, which can be conservatively kept at 0 if needed. Guidelines for achieving good results based on the theorem are brie discussed.

In this paper a genetic algorithm for clustering is proposed. The algorithm is based on the variable length chromosomes and the notion of local points density in the clustered set. Its role is to identify the number of clusters in the clustered set and to partition this set into particular clusters. The tests were conducted for two different sets of two dimensional data. The algorithm performed well in both cases. The tests presented the ability of the algorithm to partition the subsets combined with the thin dense area into separate clusters.

We consider a mathematical model which describes the adhesive contact between a linearly elastic body and an obstacle. The process is static and frictionless. The normal contact is governed by two laws. The first one is a Signorini law, representing the fact that there is no penetration between a body and an obstacle. The second one is a Winkler type law signifying that if there is no contact, the bonding force is proportional to the displacement below a given bonding threshold and equal to zero above the bonding threshold. The model leads to a variational-hemivariational inequality. We present the numerical results for solving a simple two-dimensional model problem with the Proximal Bundle Method (PBM). We analyze the method sensitivity and convergence speed with respect to its parameters.

This paper shows a new combinatorial problem which emerged from studies on an artificial intelligence classification model of a hierarchical classifier. We introduce the notion of proper clustering and show how to count their number in a special case when 3 clusters are allowed. An algorithm that generates all clusterings is given. We also show that the proposed approach can be generalized to any number of clusters, and can be automatized. Finally, we show the relationship between the problem of counting clusterings and the Dedekind problem.

The following paper discusses the topic of edge detection in X-ray hand images. It criticises the existing solution by highlighting a design fault, which is a carelessly chosen function and then proposes a way to eliminate the fault by replacing it with a better suited function. The search for this function and its results are also discussed in this paper. It also presents the aspect of pre- and postprocessing through filtering as another improvement in edge detection.

INTRODUCTION
Pattern recognition is one of principle problems in computer science. Many issues such as controlling, making decisions or predictions are related to it. It also has the main position in robotics. Therefore, this branch of computer science has been developing for a long time both in theoretical and implementation aspects. In a lot of cases pattern recognition can be a difficult problem and consequently the only method commonly used to sort out this issue does not exist. Presently, a wide range of methods based on various elements of mathematics, for instance calculus of probability or approximation theory, is applied. However, a universal recognition method does not exists - a given one can be effective for a specific sort of tasks and can fail for others. This is the reason why new methods are created and the existing ones developed. For example, syntactic methods are supported with probabilistic mechanisms and methods, which are combination of different basic methods such as neural-fuzzy ones, are created or hybrid expert systems are built. This paper concerns recognition curves in relation to their structural features. The considered problem is situated in a group of problems where pattern representation is a sequence of primitives being elements of a context language. For this group of languages, admittedly, automata which analyze these languages exist but their complexity is non-polinominal and consequently their usefulness in practical applications is limited. Moreover, an algorithm of grammar inference does not exist, consequently a method of automatic creation of tables controlling parsers (conversion functions in automata) does not exist, which in a practical nontrivial application disqualifies these languages. So, for structural patterns whose representations belong to context languages syntactic methods allowing to analyze them do not exist. Therefore, an application of nonsyntactic methods to structural features analyzing seems to be valuable. The aim of this paper is to propose a new methodology of curves recognition in relation to their structural features taking advantage of fuzzy methods statistically aided. The possibility of a neural implementation of a recognition system based on the proposed methodology is tested. In the second chapter of this paper, the methodology of a decision function construction in an axiomatic recognition of patterns is presented. In the third chapter the proposed methodology is applied to classification curves describing relative changes in the cardiac rhythm between different people with and without a cognitive load, respectively. The curves were obtained in the Department of Psychophysiology of the Jagiellonian University. The experiment is described in details in [14], [15], [27]. The fourth chapter contains the description of a neural network computing the value of membership functions for each class.

The aim of this work was to analyze the cooperation efficiency of the distributed objects based on the CORBA standard. The obtained results show the possibilities of application to the object-relational databases or object oriented computation of data received via the network from the object managed data base. The distributed objects for the client-server model were implemented in Java and C++ languages. All possible configurations of the implementation for the client and the server were analyzed. Best results were received for the client and the server implementation in C++ language. The worst results were received for the client and the server implementation in Java.

Pre-processing of mammograms is a crucial step in computer-aided analysis systems. The aim of segmentation is to extract a breast region by estimation of a breast skin-line and a pectoral muscle as well as removing radiographic artifacts and the background of the mammogram. Knowledge of the breast contour also allows further analysis of breast abnormalities such as bilateral asymmetry. In this paper we propose a fully automatic algorithm for segmentation of a breast region, based on two types of global image thresholding: the multi-level Otsu and minimizing the measure of fuzziness as well as the gradient estimation and linear regression. The results of our experiments showed that our method can be used to find a breast line and a pectoral muscle accurately

In this paper we present two methods of binarization of corneal endothelial images. The binarization is a first step of advanced image analysis. Images of corneal endothelial obtained by the specular microscopy have a poor dynamic range and they are usually non-uniformly illuminated. The binarization endothelial images is not trivial. Two binarization algorithms are proposed. The output images are presented. The quality of algorithms is discussed.

In this paper the results of numerical simulation of neurotransmitter fast transport in a presynaptic bouton of a biological neuron are presented. A mathematical model governing the transport that is fully described in [2] is recalled. A numerical simulation scheme, used parameters and their origins are described. Finally, the results of the simulation are presented.

In this paper, we place some left restrictions on derivations in CD grammar systems with phrase- structure grammars, controlled by the regular languages. The rst restriction requires that every production is always applied within the rst k nonterminals in every sentential form, for some k  1. The second restriction says how many blocks of non-terminals can be in each sentential form. The third restriction extends the second restriction and says how many blocks of non-terminals with limited length can be in each sentential form. We demonstrate that under these restrictions, the grammar systems generate dierent families of languages. Indeed, under the rst restriction, these systems generate only context-free languages. Under the second restriction, even one-component systems characterize the entire family of recursively enumerable languages. In the end, the family of languages generated by grammar systems under the third restriction is equal to the family of languages generated by programmed grammars with context-free rules without -rules of nite index.

This work is motivated by the ˇCern´y Conjecture – an old unsolved problem in the automata theory. We describe the results of the experiments on synchronizing automata, which have led us to two interesting results. The first one is that the size of an automaton alphabet may play an important role in the issue of synchronization: we have found a 5-state automaton over a 3-letter alphabet which attains the upper bound from the ˇCern´y Conjecture, while there is no such automaton (except ˇCern´y automaton C5) over a binary alphabet. The second result emerging from the experiments is a theorem describing the dependencies between the automaton structure S expressed in terms of the so-called merging system and the maximal length of all minimal synchronizing words for automata of type S.

This paper presents several normalization techniques used in handwritten numeral recognition and their impact on recognition rates. Experiments with five different feature vectors based on geometric invariants, Zernike moments and gradient features are conducted. The recognition rates obtained using combination of these methods with gradient features and the SVM-rbf classifier are comparable to the best state-of-art techniques.

A field of science such as artificial neural networks (ANN) arouses more and more interest all over the world, not only among academics and researchers. Since its beginning in the 1940‟ties it has experienced its ups and downs. From the huge fascination in the early stage to nearly falling into oblivion after Minsky‟s book [10]. Then in the 80‟ties interest in ANN experienced a rapid growth which surprised everyone and resulted in a dramatic rise and demand for information about ANN. This interest is also a result of a very dynamic development of computer science in the recent years as well as increasing need for iterative computational models, to which most definitely we can include ANN.

The idea of the project is to trace the robot position and orientation and precise it by the implementation of the advanced algorithms based on triangulation and trilateration methods. Precise navigation is a difficult task since the odometry sensors used in this context are subject to systematic and random errors. The former ones result from inaccuracies in the mechanical construction of the vehicles, e.g. the wheel diameters can be unequal or the nominal direction of the wheels is different from real.