Artificial Intelligence and Algorithms in Intelligent Systems

Advertisement is an integral part of the daily life in modern society. Nevertheless, the advertising industry often ignores the fact that people might understand ads differently and behave accordingly. Eye movements, fixation times, gaze landing sites and areas of interest in visual images are known as good predictors of customer behavior. Here we perform eye tracking of several versions (neutral, positively and negatively valence) of popular ads and analyze the dependency of eye fixation times on affective and gender factors. The results show that there are statistically significant differences on how affectively charged images are perceived, but no statistically significant differences between genders have been found when shorter fixations representing unconscious processing of visual information are analyzed.

Irrespective of a significant advancement of the compression technique in digital image processing, still the presence of artifacts or fingerprints do exists, even in a smaller scale. Such presence of artifacts is mis-utilized by the miscreants by invoking their attacks where it is quite hard to differentiate tampered image due to normal problems or malicious attack. Therefore, we present a very simple modeling of a system called as FARIP i.e. Framework of Artifact Removal in Image Processing that utilize the quantization process present in JPEG-based compression and results in perfect removal of the traces from a given image. This also acts as a solution towards the image that has been generated by the compression technique performed from JPEG standard. The comparative analysis shows that proposed system offers better signal quality as compared to the existing standards of compression.

FinFET/SRAM has been contributing to the new evolution of modern-day memory units that are used over broader scale of computing units and other sophisticated devices. A review analysis is performed over existing system to find that existing approaches are more inclined towards improvement in performance parameters and very less towards design optimization. Hence, a novel approach is introduced and is named as Search Optimization based Predictive Approach (SOPA) for optimizing the design structure of FinFET/SRAM so that it can ensure highest degree of fault tolerance when used in broader scale of dynamic applications and modern computing devices. In this, analytical methodology used where the proposed computational model is found to offer reduced computational time and more yield in increasing simulation iteration. The study contributes to progressive convergence of elite design of FinFET/SRAM rather than recursive design and hence cost effective.

Text mining is a relatively new area of computer science, and its use has grown immensely lately. The aim is to join two dataset from different data sources and to acquire information about percentage defects from the painting process, which are transmitted from the manufacturing to the end customers. The data sets are totally different and for their joining using text attributes, preprocessing are needed.

The paper considers one of the most important problem of resource allocation – packing of units within 2D space. The problem is NP-hard. A problem formulation is made, as well as restrictions and boundary conditions are found out. To solve the considered problem the authors suggest to use firefly optimization algorithm on the basis of which there are developed a bioinspired algorithm. This algorithm allows to obtain sets of quazi-optimal solutions for the 2D packing problem within polynomial time. Also, there are suggested mechanisms for encoding and decoding of alternative solutions. and presented a scheme of firefly algorithm for 2D packing problem. On the basis of the suggested algorithm there are developed software for computational experiments on benchmarks. Experimental investigations were carried out taking into account time and quality of alternative solutions. As a result, experiments shows the effectiveness of the developed algorithm.

The paper describes the results of experiments in applying the Random Projection (RP) method for authorship identification of online texts. We propose using RP for feature dimensionality reduction to low-dimensional feature subspace combined with probability density function (PDF) estimation for identification of the features of each author. In our experiments, we use the dataset of Internet comments posted on a web news site in Lithuanian language, and we have achieved 92% accuracy of author identification.

Taking gearbox as a reference structure, authors apply a method for grading the quality of mechanical structures using a convolutional neural network trained with antipatterns found in gearbox constructions. Antipatterns are used as a quality reference embodied in a neural network, which is used for classifying tested structures to match the antipatterns taught to it.The measure of similarity to antipatterns (used for training and abstracted by the neural network) is interpreted as the quality measure and so the inversed sum of similarities to each of the antipattern classes used in training is considered a quantitative grade of quality.Such grading enables automated cross-comparison of structures based on their quality (defined as differentiation from used antipatterns).

Dementia is an incurable disease that affects a large part of the population of elders and more than 21% of the elders suffering from dementia are exposed to polypharmacy. Moreover, dementia is very correlated with diabetes and high blood pressure. The medication adherence becomes a big challenge that can be approached by analyzing the daily activities of the patients and taking preventive or corrective measures. The weakest link in the pharmacy chain tends to be the patients, especially the patients with cognitive impairments. In this paper we analyze the feasibility of four classification algorithms from the machine learning library of Apache Spark for the prediction of the daily behavior pattern of the patients that suffer from dementia. The algorithms are tested on two datasets from literature that contain data collected from sensors. The best results are obtained when the Random Forest classification algorithm is applied.

Adaptive binarization of unevenly lightened images is one of the key issues in document image analysis and further text recognition. As the global thresholding leads to improper results making correct text recognition practically impossible, an efficient implementation of adaptive thresholding is necessary. The most popular global approach is the use of Otsu’s binarization which can be improved using the fast block based method or calculated locally leading to AdOtsu method. Even faster adaptive thresholding based on local mean calculated for blocks is presented in the paper. Obtained results have been compared with some other adaptive thresholding algorithms, being typically the modifications of Niblack’s method, for a set of images originating from DIBCO databases modified by addition of intensity gradients. Obtained results confirm the usefulness of the proposed fast approach for binarization of document images.

Query suggestion is an integral part of Web search engines. Data-driven approaches to query suggestion aim to identify more relevant queries to users based on term frequencies and hence cannot fully reveal the underlying semantic intent of queries. Semantic query suggestion seeks to identify relevant queries by taking semantic concepts contained in user queries into account. In this paper, we propose a machine learning approach to semantic query suggestion based on Random Forests. The presented scheme employs an optimized Random Forest algorithm based on multi-objective simulated annealing and weighted voting. In this scheme, multi-objective simulated annealing is utilized to tune the parameters of Random Forests algorithm, i.e. the number of trees forming the ensemble and the number of features to split at each node. In addition, the weighted voting is utilized to combine the predictions of trees based on their predictive performance. The predictive performance of the proposed scheme is compared to conventional classification algorithms (such as Naïve Bayes, logistic regression, support vector machines, Random Forest) and ensemble learning methods (such as AdaBoost, Bagging and Random Subspace). The experimental results on semantic query suggestion prove the superiority of the proposed scheme.

Decision making in the financial domain is a very challenging endeavor. The risk associated to this process can be diminished by gathering as much accurate and pertinent information as possible. However, most relevant data currently lies over the Internet in heterogeneous sources. Semantic Web technologies have proven to be a useful means to integrate knowledge from disparate sources. In this work, a framework to semi-automatically populate ontologies from data in semi-structured documents is proposed. The validation results in the financial domain are very promising.

The prediction of human behavior in strategic setups is important problem in many business situations. The uncertainty in human behavior complicates the problem to greater extent. It is generally assumed that players are rational in nature. But this assumption is far from actual scenarios. To address this, we propose hierarchical fuzzy deep learning network that automatically models cognitively without any expert knowledge. The architecture allows hierarchical network to generalize across different input and output dimensions by using matrix units rather than scalar units. The network’s performance is significantly better than previous models which depend on expert constructed features. The experiments are performed using datasets prepared from RPS game played over specified network and responder behavior from CT experiments. The proposed deep learning network has superior prediction performance as compared to others. The experimental results demonstrate efficiency of proposed approach.

The paper deals with the modelling of customer’s behavior in the shop of the retail chain. The paper shows that the fuzzy-expert system is a good tool for describing the behavior of a system, where the customer’s behavior is influenced by weather conditions and by events in the surroundings of the shop. The article also offers a procedure that allows dividing the system into logical units and reducing the number of necessary rules. The paper also details how the individual parts of the system have been verified. On specific real-time data the paper also presents the detection of incorrect (stereotypical) steps done by experts in compiling the knowledge base. The procedures that have been used have enabled effective identification and elimination of the errors. The advantage of our procedure was also that the IF-THEN rules that have been used were easily readable and understandable. At the end of the research work the expert system has been tested by means of available historical sales forecast data to optimize inventory, reduce storage costs, and reduce the risk of depreciation due to exceeding maximum warranty period. Achieved results have proved that fuzzy-expert systems are suitable also for the modelling of customer’s behavior and can provide us good results.

In engineering and science, there are many combinatorial optimization problems. A lot of these problems are NP-hard and can hardly be addressed by full techniques. Therefore, designing binary algorithms based on swarm intelligence continuous metaheuristics is an area of interest in operational research. In this paper we use a general binarization mechanism based on the percentile concept. We apply the percentile concept to grasshopper algorithm to solve multidimensional knapsack problem (MKP). Experiments are designed to demonstrate the utility of the percentile concept in binarization. Additionally we verify the efficiency of our algorithm through benchmark instances, showing that binary grasshopper algorithm (BGOA) obtains adequate results when it is evaluated against another state of the art algorithm.

This paper introduces the maximum likelihood estimator (MLE) based on artificial neural network (ANN) for a fast computation of the bearing that indicates the direction to the source of the electromagnetic wave received by a passive radar system equipped with an array antenna. Authors propose the cascade scheme for ANN training phase where the network is fed with the pair-wise delays of received stationary or cyclostationary signals and the output of the network goes to the input of the target function being maximized together with the same data. The designed ANN topology has the modified output layer consisting of the custom neuron that implements argument function of a complex number rather than linear or sigmoid-like ones used in the conventional multilayer perceptron topologies. The simulation carried out for the ring array antenna shows that a single estimation obtained via ANN MLE takes 12 times less computational time comparing to the MLE implemented via the numerical optimization technique. The degradation of accuracy measured as the increase of mean-squared error does not exceed 10% of the potential value for the particular signal-to-noise ratio (SNR) and that difference has no tendency to decrease for higher SNR. The estimation error appeared to be independent from the true value in the wide range of bearings.

The paper presents approach to knowledge discovery by using query expansion to search for cross-lingual mathematical terminology extraction. It employs information retrieval and statistically-based techniques to extract and process keyword collocations in large comparable cross-lingual web electronic text corpora in the domain of mathematics in Bulgarian and in Serbian language. It, also, offers examples and survey of used techniques for semantic search and clustering by comparing keyword collocations to build a cross-lingual thesauri. The results of semantic keyword search for the two web electronic text corpora using Sketch Engine software are presented and analyzed with respect to the types of keyword collocations processing and to multilingual application of the approach.

Search engine optimization involves various techniques which web site creators and marketers can apply on web pages with goal of ranking higher in popular search engines. One of important ranking factors is “meta description” - a short textual description of the website which is put inside web page header accessible to web spiders. In this paper we investigate if existing text summarization techniques can be used to artificially build “meta description” for websites which are missing it. Also, we propose a simple query based algorithm for generation of “meta description” content based on some summarization techniques. The experimental results and expert evaluation show that proposed algorithm for text summarization can successfully be used in this context. Results of this research can be used to build recommender system for improvement of search engine optimization of a webpage.

The paradigm of the swarm algorithm is proposed on the basis of integration of models of adaptive behavior of ant and bee colony. Integration of models is reduced to the creation of a hybrid agent alternately performing the functions of adaptive behavior of ant and bee colony. The proposed class of hybrid algorithms can be used to solve a wide range of combinatorial problems Based on the hybrid paradigm, a partitioning algorithm has been developed. Also article give a comparison hybrid algorithms with other methods of solution problem. Compared with the existing algorithms, the improvement of results is achieved by 5–10%. The probability of obtaining the global optimum was 0.9.

This paper presents heuristics approach for the problem of generation an optimal multistage tourist route of electric vehicle (EV). For the given starting and a final point (being EV charging stations) the points of interests (POIs) are included which maximizing the tourist attractiveness. Furthermore the intermediate EV charging stations are selected to the route, in order to after specified number of kilometers a tourist could recharge the batteries and move on to the next stage of a route. Greedy algorithm strengthened the local search methods is proposed by us. Computational tests are conducted on realistic database including POIs and EV charging stations. Results and the execution time of the algorithm show that the presented solution could be a part of software module which generates the most interesting route taking into consideration driving range of EV battery.

Sampling technique has become one of the recent research focuses in the graph-related fields. Most of the existing graph sampling algorithms tend to sample the high degree or low degree nodes in the complex networks because of the characteristic of scale-free. Scale-free means that degrees of different nodes are subject to a power law distribution. So, there is a significant difference in the degrees between the overall sampling nodes. In this paper, we propose a concept of approximate degree distribution and devise a stratified strategy using it in the complex networks. We also develop two graph sampling algorithms combining the node selection method with the stratified strategy. The experimental results show that our sampling algorithms preserve several properties of different graphs and behave more accurately than other algorithms. Further, we prove the proposed algorithms are superior to the off-the-shelf algorithms in terms of the unbiasedness of the degrees and more efficient than state-of-the-art FFS and ES-i algorithms.

Clustering algorithms often use distance measure as the measure of similarity between point pairs. Such clustering algorithms are difficult to deal with the curse of dimensionality in high-dimension space. In order to address this issue which is common in clustering algorithms, we proposed to use MIC instead of distance measure in k-means clustering algorithm and implemented the novel MIC-kmeans algorithm for high-dimension clustering. MIC-kmeans can cluster the data with correlation to avoid the problem of distance failure in high-dimension space. The experimental results over the synthetic data and real datasets show that MIC-kmeans is superior to k-means clustering algorithm based on distance measure.

Data exploration has been proved to be an efficient solution to learn interesting new insights from dataset in an intuitional way. Typically, discovering interesting patterns and objects over high-dimensional dataset is often very difficult due to its large search space. In this paper, we developed a data exploration method named Decision Analysis of Cross Clustering (DACC) based on subspace clustering. It characterize the data objects in the representation of decision trees over divided clustering subspace, which help users quickly understand the patterns of the data and then make interactive exploration easier. We conducted a series of experiments over the real-world datasets and the results showed that, DACC is superior to the representative data explorative approach in term of efficiency and accuracy, and it is applicable for interactive exploration analysis of high-dimensional data sets.

In order to surmount the major difficulties in multi-target tracking, one was that the observation model and target distribution was highly non-linear and non-Gaussian, the other was varying number of targets bring about overlapping complex interactions and ambiguities. We proposed a kind of system that is able of learning, detecting and tracking the multi-targets. In the method we combine the advantages of two algorithms: mixture particle filters and Multiple Instance Boosting. The key design issues in particle filtering are the selecting of the proposal distribution and the handling the problem of objects leaving and entering the scene. We construct the proposal distribution using a compound model that incorporates information from the dynamic models of each object and the detection hypotheses generated by Multiple Instance Boosting. The learned Multiple Instance Boosting proposal distribution makes us to detect quickly object which is entering the scene, while the filtering process allows us to keep the tracking of the simple object. An automatic multiple targets tracking system is constructed, and it can learn and detect and track the interest object. Finally, the algorithm is tested on multiple pedestrian objects in video sequences. The experiment results show that the algorithm can effectively track the targets the number is changed.

Number of existing signal processing methods can be used for extracting useful information. However, receiving desired and eliminating undesired information is yet a significant problem of these methods. Empirical Mode Decomposition (EMD) algorithm shows promising results in comparison to other signal processing methods especially in terms of accuracy. For example, it shows an efficient relationship between signal energy and time frequency distribution. Though, EMD algorithm still has a noise contamination which may compromise the accuracy of the signal processing. It is due to the mode mixing phenomenon in the Intrinsic Mode Function’s (IMF) which causes the undesirable signal with the mix of additional noise. Therefore, it has still a room for the improvements in the selective accuracy of the sensitive IMF after decomposition that can influence the correctness of feature extraction of the oxidized carbon steel. This study has used two datasets to compare the parameters analysis of the Ensemble Empirical Mode Decomposition (EEMD) algorithm for constructing the signal signature.

An important factor in the recognition of magnetic resonance images is not only the accuracy, but also the speed of the segmentation procedure. In some cases, the speed of the procedure is more important than the accuracy and the choice is made in favor of a less accurate, but faster procedure. This means that the segmentation method must be fully adaptive to different image models, that reduces its accuracy. These requirements are satisfied by developed hyper-heuristical particle swarm method for image segmentation. The main idea of the proposed hyper-heuristical method is the application of several heuristics, each of which has its strengths and weaknesses, and then their use depending on the current state of the solution. Hyper-heuristical particle swarm segmentation method is a management system, in the subordination of which there are three bioinspired heuristics: PSO-K-means, Modified Exponential PSO, Elitist Exponential PSO. Developed hyper-heuristical method was tested using the Ossirix benchmark with magnetic-resonance images (MRI) with various nature and different quality. The results of method’s work and a comparison with competing segmentation methods are presented in the form of an accuracy chart and a time table of segmentation methods.

The research of EEG classification is of great significance to the application and development of brain-computer interface. The realization of brain-computer interface depends on the good accuracy and robustness of EEG classification. Because the brain electrical capacitance is susceptible to the interference of noise and other signal sources (EMG, EEG, ECG, etc.), EEG classifier is difficult to improve the accuracy and has very low generalization ability. A novel method based on sparse autoencoder (SAE) and convolutional neural network (CNN) is proposed for feature extraction and classification of motor imagery electroencephalogram (EEG) signals. The performance of the proposed method is evaluated with real EEG signals from different subjects. The experimental results show that the network structure can get better classification results than other classification algorithms.

Web search engines are successful at finding relevant resources; however, the search engine results page (SERP) list contains so many results that the intended ones are difficult to identify, therefore, requires manual user inspection and selection. A user may equally likely visit an already visited web page, which requires them to repeat the same search process. To deal with this issue, several re-visitation approaches are used, which include history, bookmarking and URL auto-completion. Among these, bookmarking is the most effective and user-friendly approach. Today, bookmarking is available in almost all the frequently used web browsers. However, they provide static and unstructured bookmarks. This makes it difficult for the end-user to easily manage, organize and maintain the hierarchical structure of bookmarks especially when their number exceeds a certain limit. Besides their hierarchical structure, the available bookmarking systems provide keyword-based searching with no exploitation of the semantics of the visited resources making it difficult to re-visit a required resource without reference to its context. We exploit Semantic Web technologies to devise a more effective, accurate and precise bookmarking service so that the cognitive efforts of the users could be reduced to a significant level. The proposed solution uses an extension that uses ontology to generate semantic bookmarks by tracking the user browsing activities, resulting into a more user-friendly re-visitation experience.

A model complex and algorithms for assessing the technical condition (TC) and reliability of the launch vehicle (LV) “Soyuz-2” with the decision support (DS) for managing its life cycle (LC) is considered in the article. On the basis of the analysis of modern problems and the requirements for the efficiency, quality and reliability of the assessment of the LV and the reliability of LV, it was concluded that it is necessary to use the new intelligent information technology (IIT), presented in the article, when designing the automated monitoring systems of the condition and DS for managing the LC LV “Soyuz-2”. As a theoretical basis for this technology, the modification of the generalized computational model (GCM) as a knowledge representation model, allowing to build simulation-analytical model-based complexes for monitoring conditions and managing complex organizational and technical objects (COTO), is considered.

The article describes the approach to designing proactive decision support systems (DSS) based on the modification of the generalized computational model as a knowledge representation model (KRM). It is proposed to use a precedent methodology for making management decisions. It was found that this approach allows to predict the situations and ensure the model’s self-learning by forming a precedent knowledge base.

The possibility of creating thinking systems discusses issues that may arise in the near future of AI. However this outlines challenges to ensure that AI operates safely as it approaches humans in its intelligence from Algorithms Intelligent Systems. To understand how progress may proceed we need to understand how existing algorithms are developed and improve, differentiating the concepts between data analytics and data algorithmic decision making. This article reviews the literature on AI and AIS and presents some general guidelines and a brief summary of research progress and open research questions. The first section reviews the basic foundation of Artificial Intelligence to provide a common basis for further discussions and the second section of this paper suggests the development of Algorithm Intelligence Systems models including: learning algorithms such as learning from observations, learning in Neural and Belief Networks and reinforcement learning.

The article deals with a way of automated evaluation of competent persons evaluated as a percentage. In the introduction, the need of evaluating professional competencies of new or existing employees is outlined. We define the problem of overall evaluating various professional competencies. Subsequently, we propose a hierarchical system that automatically generates aggregate evaluation of any number of evaluating computations including continuous visualisation of the results. As a calculating unit, we use the Takagi-Sugeno expert system. This solution can be extended with other attributes that can be evaluated for competencies. The proposed hierarchical system is experimentally verified.

The article presents the data mining of dataset about spreading information among the university study applicants. The data were collected during admission procedure of applicants for bachelor study branch Geoinformatics and Geography at Palacky University in Olomouc (Czech Republic). Answers were received by questionnaire in two years, 2016 and 2017. Data collecting and processing aimed to discover the dissemination of first information about this specific specialization among graduates at secondary schools. Statistics and data mining techniques, namely finding association rule were used. Data mining discovered some unexpected relation and association in the data. Interesting results bring feedback about the impact of various presentation activities like Open Day, GIS Day or publishing information on the Internet. Results will also be reflected in future advertisement strategies of the study branch Geoinformatics to assure increasing interest of the potential applicants.

Keystroke dynamics is investigated over 30 years because of its biometric properties, but most of the studies are focusing on identification. In current study our goal is to predict user age by keystroke data. We collected keystroke data through different real life online systems during 2011 and 2018. Data logs were labeled with user age, gender and in some cases with other available information. We analyzed 2.3 million keystrokes, from 7119 keystroke data logs, produced by ca 1000 individual subjects, presenting six different age groups. All these data logs are also made available to research community, and the web address is provided in the paper. We carried out binary and multiclass classification using supervised machine-learning methods. Binary classification results were all over the baseline, best f-score over 0.92 and lowest 0.82. Multiclass classification distinguished all groups over baseline. Analyzing distinguishing features, we found overlap with text-mining features from previous studies.

It is quite challenging for the human resource management professionals to deal with the employees’ salary expectations compared to their work during the salary increment period. Employees’ salary is raised based on some major factors which include their mastery, responsibility, workload factors and many more. For the salary increment, the employee’s performance is estimated based on crisp values. This estimation leads to uncertainty and vagueness. In order to handle this situation of uncertainty, we have proposed the use of fuzzy logic in salary increment model. Our data set consists of the salary increment factors of the employees along with the increased salary percentage of 100 employees. The implementation of Adaptive Neuro Fuzzy inference system (ANFIS) on salary increment model is described in this paper. We have approached the Sugeno fuzzy inference model to generate the fuzzy rules and membership functions of input (salary increment factors) and output (salary increment percentages) data.

Human race is affected by a silent but dangerous disease known as obesity. This is the result of the difference between the high calorie intake and low calorie burn. Though most of the people do not realize that they are slowly being engulfed by this disease which can be easily prevented by some simple measures. Along with obesity, underweight is also an issue. To avoid these issues a calorie based diet is the best solution. To recover from this obesity, people need to lose weight and to recover from underweight, people need to gain some weight, both depend on calorie intake and calorie burn. So, to assist with this weight-loss and weight-gain a fuzzy logic based weight loss/gain training program is proposed in this research.

The paper demonstrates the use of clustering to find different sensitive seismic zones and time series for the earthquake hazard prediction. Anticipating seismic activities using previous history data is obtained by applying hierarchical, k-means and density based clustering. Data is collected first and then clustered. Finally, the clustered data is used to obtain the different seismic zones on map. On the top of that data is used in linear regression to build a predictive model for forecasting upcoming earthquakes’ magnitudes for different regions in and nearby areas of Bangladesh.

In heterogeneous data processing, various data model often make analytic task too hard to achieve optimal performance, it is necessary to unify heterogeneous data into the same data model. How to determine the proper intermediate data model and unify the involved heterogeneous data models for the analytical task is an urgent problem need to be solved. In this paper, we proposed a model determination method based on cost estimation. It evaluates the execution cost of query tasks on different data models, which taken as the criterion to measure the data model, and chooses a data model with the least cost as the intermediate representation during data processing. The experimental results of BigBench datasets showed that the proposed cost estimation based method could appropriately determine the data model, which made heterogeneous data processing efficiently.

This paper describes architecture of an artificial intelligence system based on the Elman neural network. Simple training algorithms and neural network models are not able to solve such a complex problem as movements in the conditions of an independent game world environment, so a combination of a base neural network training algorithm and Q-learning agent approach is used as part of a player behavior control model. The paper also includes results of experiments with different values of model and game world characteristics and shows efficiency of the described approach.

Neural network-based word embeddings –outperforming traditional approaches in the various Natural Language Processing tasks – have gained a lot of interest recently. Despite it, the Lithuanian word embeddings have never been obtained and evaluated before. Here we have used the Lithuanian corpus of $$\sim $$234 thousand running words and produced several word embedding models: based on the continuous bag-of-words and skip-gram architectures; softmax and negative sampling training algorithms; varied number of dimensions (100, 300, 500, and 1,000). Word embeddings were evaluated using the Lithuanian WordNet as the resource for the synonym search. We have determined the superiority of the continuous bag-of-words over the skip-gram architecture; while the training algorithm and dimensionality showed no significant impact on the results. Better results were achieved with the continuous bag-of-words, negative sampling and 1,000 dimensions.

In the paper a texture classification method utilizing the Gray Level Co-occurrence Matrix (GLCM) is proposed which can be applied for autonomous cleaning robots. Our approach is based on the analysis of chosen Haralick features calculated locally together with their selected statistical properties allowing to determine the additional features used for classification purposes. To verify the presented approach a dedicated color image dataset containing textures selected from the Amsterdam Library of Textures (ALOT) representing surfaces typical for the autonomous cleaning robots has been used. The results obtained for various color models and three different classifiers confirm the influence of the color model as well as the advantages of the proposed extended GLCM based approach.

The task of operational planning of production is considered. The hierarchy of tasks of production planning is described. The formulation of the problem in terms of scheduling theory is given. A model for solving the problem of operational planning as an adaptive system is proposed. The software agent architecture is chosen, local and global goals of the adaptive system are formed. The use of the apparatus of fuzzy sets in the determination of the agent state is justified. Computational experiments were carried out and the results obtained were analyzed.

Stacked autoencoder was used for the segmentation of trabeculas from bone marrow histological images derived from patients after hip joints arthroplasty. Additional filtering of areas smaller than 20000 pixels is necessary. The method has 95% efficiency. Proposed stacked autoencoder processes input images without special intervention automatically that is the main advantage of unsupervised learning over the supervised learning.

Several scholarly social networking platforms are available on the Web, which build a collaborative research & development (R&D) environment for academicians and researchers to connect and collaborate with each other in solving real-world problems. The collaboration happens in the form of uploading, sharing and following research outcomes including technical reports, research publications, books, etc.; giving feedback on these research outputs; and community-based question & answering (CQA). In such systems, the reputation of users plays a key role, which acts as a trust indicator for the quality of questions, answers, and in recommending scholars and scholarly data. It is therefore necessary to build the reputation of a scholar in manner that reflects their active participation in the CQA activities. Therefore, the paper contributes a reputation model that besides expertise, considers the willingness of the user to participate in the CQA activities. The proposed reputation model is the first step towards recommending experts and active scholars that can potentially answer a given question. The empirical results show that the user expertise and their willingness to participate in the scholarly social Q&A activities play a major role in building more accurate reputation.

This paper studies the performance of a newly developed optimization algorithm inspired by the behavior of bison herds: the Bison Algorithm. The algorithm divides its population into two groups. The exploiting group simulates the swarming behavior of bison herds endangered by predators. The exploring group systematically runs through the search space in order to avoid local optima.At the beginning of the paper, the Bison Algorithm is introduced. Then the performance of the algorithm is compared to the Particle Swarm Optimization and the Cuckoo Search on the set of 30 benchmark functions of IEEE CEC 2017. Finally, the outcome of the experiments is discussed.

This work studied a relationship between optimization qualities of Success-History based Adaptive Differential Evolution algorithm (SHADE) and its self-adaptive parameter strategy. Original SHADE with improvement based adaptation is compared to the SHADE with Distance based parameter adaptation (Db_SHADE) on the basis of the CEC2015 benchmark set for continuous optimization and a novel approach combining both distance and improvement adaptation (DIb_SHADE) is presented and tested as a trade-off between both approaches.

The paper deals with an approach for a reliable dogface detection in an image using the convolutional neural networks. Two detectors were trained on a dataset containing 8351 real-world images of different dog breeds. The first detector achieved the average precision equal to 0.79 while running real-time on single CPU, the second one achieved the average precision equal to 0.98 but more time for processing is necessary. Consequently, the facial landmark detector using the cascade of regressors was proposed based on those, which are commonly used in human face detection. The proposed algorithm is able to detect dog’s eyes, a muzzle, a top of the head and inner bases of the ears with the 0.05 median location error normalized by the inter-ocular distance. The proposed two-step technique – a dogface detection with following facial landmark detector - could be utilized for a dog breeds identification and consequent auto-tagging and image searches. The paper demonstrates a real-world application of the proposed technique – a successful supporting system for taking pictures of dogs facing the camera.

Orthogonal learning strategy, a proven technique, is combined with hybrid optimization metaheuristic, which is based on Firefly Algorithm and Particle Swarm Optimization. The hybrid algorithm Firefly Particle Swarm Optimization is then compared, together with canonical Firefly Algorithm, with the newly created Orthogonal Learning Firefly Algorithm. Comparisons have been conducted on five selected basic benchmark functions, and the results have been evaluated for statistical significance using Wilcoxon rank-sum test.

This paper provides a closer insight into applicability and performance of the hybridization of symbolic regression open framework, which is Analytical Programming (AP) and Differential Evolution (DE) algorithm in the task of time series regression. AP can be considered as a robust open framework for symbolic regression thanks to its usability in any programming language with arbitrary driving metaheuristic. The motivation behind this research is to explore and investigate the applicability and differences in performance of AP driven by basic canonical entirely random or best solution driven mutation strategies of DE. An experiment with four case studies has been carried out here with the several time series consisting of GBP/USD exchange rate. The differences between regression/prediction models synthesized using AP as a direct consequence of different DE strategies performances are statistically compared and briefly discussed in conclusion section of this paper.