We advance our work on a special text categorization problem, the multiaspect text categorization, introduced in our previous works. In general case, it assumes a hierarchy of categories, and documents are assigned to leaves of a category but within categories documents are further structured into sequences of documents, referred to as cases. This is much more complex than the classic text categorization. Previously, we proposed a number of approaches to deal the above problem but we took into account to a limited extent hierarchies occurring in the definition of the problem. Here, we we start with one of our best approaches proposed so far and extend it by assuming that categories are arranged into a hierarchy, and that there is a hierarchical relation between a category and its offspring cases.
The paper presents experimental evaluation of image classification in the field of web content filtering using bag of visual features and convolutional neural networks approach. A more difficult data set than traditionally used ones was built from very similar types of images in order to make conditions closer to real world practice. F1-measure of classifiers that are based on bags of visual features was significantly lower than that reported in previously published papers. Convolutional neural networks performed much better. Also, we measured and compared training and prediction time of various algorithms.
This paper focuses on a statistical comparison with a proposed Fuzzy BCO based on an Interval Type-2 Fuzzy System and the Original BCO algorithm using Trapezoidal Membership Functions. The Fuzzy Bee Colony Optimization method applied for tuning the parameters of the Fuzzy Logic Controller is presented. The objective of the work is based on the main reasons for the statistical comparison of BCO and Fuzzy BCO algorithm that is to find the optimal design in the fuzzy logic controller for two problems in fuzzy control. We added perturbations in the model with band-limited noise so that the Interval Type-2 Fuzzy Logic System is better analyzed under uncertainty and to verify that the Fuzzy BCO shows better results than the Original BCO.
Modern industries and business firms are widely using data mining applications in which the problem of Frequent Itemset Mining (FIM) has a major role. FIM problem can be solved by standard traditional algorithms like Apriori in certain transactional database and can also be solved by different exact (UApriori, UFP Growth) and approximate (Poisson Distribution based UApriori, Normal Distribution based UApriori) probabilistic frequent itemset mining algorithm in uncertain transactional database (database in which each item has its existential probability). In our algorithm it is considered that database is distributed among different locations of globe in which one location has certain transactional database, we call this location as main site and all other locations have uncertain transactional databases, we call these locations as remote sites. To the best of our knowledge no algorithm is developed yet which can calculate frequent itemsets on the combination of certain and uncertain transactional database. We introduced a novel approach for finding itemsets which are globally frequent among the combination of all uncertain transactional databases on remote site with certain database at main site.
The game approaches are rather popular in many applications, where a collective of automata is used. In the present paper such a collective consists in a group of learning automata characterized with simple number parameters. The fuzzy measuring implemented as a game which is played sequentially with one automaton at a time, the result of the game defines next automaton to be played with. This game provides some measuring system that is very close to the procedure of collecting statistics in Probability Theory. For measuring of an unknown membership function a new concept has been introduced called Cognitive Generator which transforms a fuzzy singleton to ordinary crisp logic value. Considerations on various types of axiomatic approaches shows that the Cognitive Generator, as well as the Evidence Combination Axiomatic, belongs to one class of axiomatic theories, which may be used in application directly. The present paper contains also some programming examples aimed to illustrate our general approach.
In the traditional chemical kinetics, the rate of each reaction A + ... + B -> ... is proportional to the product c(A) center dot ... center dot c(B) of the concentrations of all the input substances A,..., B. For high concentrations c(A), ... , c(B), the reaction rate is known to be proportional to the minimum min(c(A), ... , c(B)). In this paper, we use fuzzy-related ideas to derive the formula of the reaction rate for situations intermediate between usual and high concentrations.
In this paper, we show how fuzzy and probabilistic techniques can be used in environment-related data processing. Specifically, we will show that these methods help in solving two environment-related problems: how to predict the birds' nesting sites and how to measure shoreline erosion.
In this paper an ensemble of three neural networks with type-2 fuzzy weights is proposed. One neural network uses type-2 fuzzy inference systems with Gaussian membership functions for obtain the fuzzy weights; the second neural network uses type-2 fuzzy inference systems with triangular membership functions; and the third neural network uses type-2 fuzzy inference systems with triangular membership functions with uncertainty in the standard deviation. Average integration and type-2 fuzzy integrator are used for the results of the ensemble neural network. The proposed approach is applied to a case of time series prediction, specifically in the Mackey-Glass time series.
In principle, distributed heterogeneous commodity clusters can be deployed as a computing platform for parallel execution of user application, however, in practice, the tasks of first discovering and then configuring resources to meet application requirements are difficult problems. This paper presents a general-purpose resource selection framework that addresses the problems of resources discovery and configuration by defining a resource selection scheme for locating distributed resources that match application requirements. The proposed resource selection method is based on the frequencies of weighted condition attribute values of resources and the outstanding overall searching ability of genetic algorithm. The concept of soft set condition attributes reducts, which is dependent on the weighted conditions' attribute value of resource parameters is used to achieve the required goals. Empirical results are reported to demonstrate the potential of soft set condition attribute reducts in the implementation of resource selection decision models with relatively higher level of accuracy.
This work describes the idea of an adaptive semantic layer for large-scale databases, allowing to effectively handling a large amount of information. This effect is reached by providing an opportunity to search information on the basis of generalized concepts, or in other words, linguistic descriptions. These concepts are formulated by the user in natural language, and modelled by fuzzy sets, defined on the universum of the significances of the attributes of the database. After adjustment of user's concepts based on search results, we have "personalized semantics" for all terms which particular person uses for communications with database (for example, "young person" will be different for teenager and for old person; "good restaurant" will be different for people with different income, age, etc.).
In the paper, the application of Mamdani-type fuzzy inference method to the expert evaluation of the impact of tax administration reforms on the tax potential is investigated. As input data of the system are taken reforms in tax administration and fuzzified by the triangle, trapezoid, Gaussian and Bell membership functions. It has been shown that the suggested fuzzy approach is one of the effective methods for evaluation of tax potential.
Urban air quality has degraded at an alarming rate due to rapid urbanisation and industrialization in megacities. Therefore, there is an urgent need to assess air quality and suggest risk mitigation measures. In this paper, air quality of Chennai city was evaluated using different Fuzzy Synthetic Evaluation (FSE) techniques i.e. Fuzzy similarity method (FSM) and Simple fuzzy classification (SFC) and the results are compared with the National air quality index (NAQI). In the case of SFC weights for different pollutants were computed using Shannon's information entropy. Seasonal analysis of the criteria pollutants shows highest concentration during the winter season followed by pre-monsoon and summer season. The lowest concentration was observed during Monsoon in most cases. The FSE results are optimistic as compared to the NAQI due to aggregation of pollutant concentration as opposed to maximising function in NAQI which reconfirms the findings of earlier researchers. FSE can be used as a decision making tool to communicate the overall air quality to policy makers/end users (Public) in a simplified qualitative form.
In this work, we applied Adaptive Neuro-Fuzzy Inference System to three different classification problems: (1) sentence-level subjectivity detection, (2) sentiment analysis of texts, and (3) detecting user intention in natural language call routing system. We used English dataset for the first and second problems, but Azerbaijani dataset for the third problem based on same features. Our feature extraction algorithm calculates a feature vector based on the statistical occurrences of words in a corpus without any lexical knowledge.
Lukasiewicz fuzzy systems are fuzzy systems based on Lukasiewicz implication and Lukasiewicz t-norm and t-conorm as fuzzy operations. They are deeply rooted in classical logic while being fuzzy systems, so they establish a connection between classical logic and fuzzy logic. Lukasiewicz fuzzy systems with Center of Gravity defuzzification have been shown to have good approximation properties, however Center of Gravity defuzzification makes them to be computationally not very efficient. In the present paper we develop a real-time Lukasiewicz fuzzy system, using the Mean of Maxima defuzzification. This defuzzification will be directly computable for Lukasiewicz systems with certain properties. We investigate approximation properties of such systems and we obtain a generalization of a previous universal approximation result.
Authors measured electroencephalograms (EEGs) as participants recognized and recalled 13 playing card images (from ace to king of club) presented on a CRT monitor. During the experiment, electrodes were fixed on the scalps of the participants. Four EEG channels located over the right frontal and temporal cortices (Fp(2), F-4, C-4 and F-8 according to the international 10-20 system) were used in the discrimination. Sampling data were taken at latencies between 400 and 900 ms at 25 ms intervals for each trial. Thus, data were 84 dimensional vectors (21 time point X 4 channels). The number of objective variables was 13 (the number of different cards), and the number of explanatory variates was thus 84. Canonical discriminant analysis was applied to these single trial EEGs. Results of the canonical discriminant analysis were obtained using the jack knife method and were 100% of nine participants. We could perform playing card estimation magic without a trick. This fact is sub production based on our series of precedent research.
Heating, ventilation and air-conditioning (HVAC) system is an important component of Smart Home. The HVAC system is connected to network for the transfer of measurement data and control action packets from sensors to controller and controller to actuators respectively. The HVAC system can therefore be categorized as a Cyber-Physical system (CPS). Such systems are prone to communication uncertainties like packet losses and delays. Such systems require integrated architecture of communication and control. An evolutionary algorithm tuned fuzzy PI controller design coupled in a communication framework is presented in this paper for performance improvements of HVAC system. The entire architecture considers relevant system objectives based on system states and actuator actions. The formulated problem has been solved through real time optimization approach using the designed controller following the communication protocol. The developed algorithm helps in obtaining optimal actuator actions and shows a fast convergence to the different desired temperature sets. The results also show that the system can recover from sudden burst packet losses.
Minimum Weight Dominating Set (MWDS) belongs to the class of NP-hard graph problem which has several real life applications especially in wireless networks. In this paper, we present a new hybrid genetic algorithm. Also, we propose a new heuristic algorithm for MWDS to create initial population. We test our hybrid genetic algorithm on (Jovanovic et al., Proceedings of the 12th WSEAS international conference on automatic control, modeling and simulation, 2010) [3] data set. Then the results are compared with existing algorithms in the literature. The experimental results show that our hybrid genetic algorithm can yield better solutions than these algorithms and faster than these algorithms.
There exist different methods of malware identification, while the most common is signature-based used by anti-virus vendors that includes one-way cryptographic hash sums to characterize each particular malware sample. In most cases such detection results in a simple classification into malware and goodware. In a modern Information Security society it is not enough to separate only between good ware and malware. The reason for this is increasingly complex functionality used by various malware families, in which there has been several thousand of new ones created during the last decade. In addition to this, a number of new malware types have emerged. We believe that Soft Computing (SC) may help to understand such complicated multinomial problems better. To study this we ensambled a novel large-scale dataset based on 400 k malware samples. Furthermore, we investigated the limitation of community-accepted Soft Computing methods and can clearly observe that the optimization is required for such non-trivial task. The contribution of this paper is a thorough investigation of large-scale multinomial malware classification by Soft Computing using static characteristics.
Hierarchical fuzzy systems are one of the most popular solutions for the curse of dimensionality problem occurred in complex fuzzy rule based systems with a large number of input parameters. Nevertheless these systems have a hidden inaccuracy and instability problem. In detail, the outputs of hierarchical systems, based on Mamdani style inference, differ from the outputs of equivalent single system. Moreover they are not stable in any variation of system modeling even if the rules and membership functions do not expose any differentiation. This paper revisits inaccuracy and instability problems of hierarchical fuzzy inference systems. It investigates the differentiation in systems’ behaviors against the variations in system modeling, and provides a pattern to identify the magnitude of this differentiation.
