Recent Advances in Soft Computing

The constraint satisfaction problem is a useful and well-studied framework for the modeling of many problems rising in Artificial Intelligence and other areas of Computer Science. As many real-world optimization problems become increasingly complex and hard to solve, better optimization algorithms are always needed. Genetic algorithms (GA) which belongs to the class of evolutionary algorithms is regarded as a highly successful algorithm when applied to a broad range of discrete as well continuous optimization problems. This paper introduces a hybrid approach combining a genetic algorithm with the multilevel paradigm for solving the maximum constraint satisfaction problem. The promising performances achieved by the proposed approach is demonstrated by comparisons made to solve conventional random benchmark problems.

In the queueing theory, it is assumed that requirement arrivals correspond to the Poisson process and the service time has the exponential distribution. Using these assumptions, the behaviour of the queueing system can be described by means of the Markov chains and it is possible to derive characteristics of the system. In the paper, these theoretical approaches are presented and focused on systems with several service lines and the FIFO queue when the number of requirements exceeds the number of lines. Finally, it is also shown how to compute the characteristics in a situation when these assumptions are not satisfied.

The presented paper deals with problem of studying of Random Number Generators onto properties (especially efficiency) of GPA-ES algorithm. The first chapter brings simple introduction into the problem followed by description of GPA-ES algorithm from the viewpoint of influence of RNGs onto its function. The third chapter then describes organization of experiments and the next one brings their implementation details and simulation results. Then the fifth chapter discusses results, especially influence of different RNGs onto GPA-ES algorithm function and the sense and form on the next experiments.

This paper investigates the possibility of creating ontology concepts from information contained in a database, by finding random queries with the help of a genetic algorithm. This is done, with the aim to help ontology building. Based on the structure of the database random chromosomes are created. Their genes describe possible selection criteria. By using a genetic algorithm, these selections are improved. Due to the size of the database, an approach for finding fitness from general characteristics, instead of an in-depth analysis of the data is considered. After the algorithm finished improving the chromosomes in the population, the best chromosomes are chosen. They are considered for implementation as ontology concepts. These ontology concepts can be used as descriptions of the information contained in the database. Because genetic algorithms are not usually used for ontology building, this paper investigates the feasibility of such an approach.

Artificial fish swarm algorithm is a technique based on swarm behaviors that are inspired from schooling behaviors of fishes swarm in the nature. Group escaping is another interesting behavior of fish that is ignored. This behavior shows all fish change their moving directions rapidly while some fish sense a predator. In this paper, we proposed a new algorithm which is obtained by hybridizing artificial fish swarm algorithm and group escaping behavior of fish which can greatly speed up the convergence. It is presented proper pseudocode of improved algorithm and then experimental results on Traveling Salesman Problem is applied and demonstrated the advantages of the improved algorithm.

There are many optimization algorithms which can be used for solving different tasks. One of those is the genetic programming method, which can build an analytical function which can describe data. The function is coded in a tree structure. The problem is that when we decide to use lower maximal depth of the tree, the genetic programming is not able to compose a function which is good enough. This paper describes the way how to solve this problem. The approach is based on creating partial solutions represented by subtrees and composing them together to create the last tree. This approach was tested for finding a function which can correctly calculate the output according to the given inputs. The experiments showed that even when using a small maximal depth, the genetic programming using our approach can create functions with good results.

Linear genetic programming (LGP) represents candidate programs as sequences of instructions for a register machine. In order to accelerate the evaluation time of candidate programs and reduce the overall time of evolution, we propose various parallel implementations of LGP suitable for the current multi-core processors. The implementations are based on a parallel evaluation of candidate programs and the island model of the parallel evolutionary algorithm in which the subpopulations are evolved independently, but some genetic material can be exchanged by means of the migration. Proposed implementations are evaluated using three symbolic regression problems and a hash function design problem.

This research deals with the hybridization of several approaches for evolutionary algorithms, which are the adaptive control parameter adjusting strategy and multi-chaotic dynamics driving the selection of indices in Differential Evolution (DE). The novelty of the paper is given by the experiments with the multi-chaos-driven adaptive DE concept inside adaptive parameter adjusting DE strategies. These experiments are representing the investigations on the mutual influences of several different randomizations types together with adaptive DE strategies. The multi-chaotic concept is representing the adaptive switching between two different chaotic systems based on the progress of individuals within population. This paper is aimed at the embedding of discrete dissipative chaotic systems in the form of multi-chaotic pseudo random number generators for the jDE, which is the state of the art representative of simple adaptive control parameter adjusting strategy for DE. Repeated simulations for two different combinations of driving chaotic systems were performed on the IEEE CEC 13 benchmark set. Finally, the obtained results are compared with the canonical not-chaotic jDE.

Researchers in artificial intelligence usually adopt the Satisfiability paradigm as their preferred methods when solving various real worlds decision making problems. Local search algorithms used to tackle different optimization problems that arise in various fields aim at finding a tactical interplay between diversification and intensification to overcome local optimality while the time consumption should remain acceptable. The WalkSAT algorithm for the Maximum Satisfiability Problem (MAX-SAT) is considered to be the main skeleton underlying almost all local search algorithms for MAX-SAT. This paper introduces an enhanced variant of WalkSAT using Finite Learning Automata. A benchmark composed of industrial and random instances is used to compare the effectiveness of the proposed algorithm against state-of-the-art algorithms.

Large scale optimization is challenging area due to the curse of dimensionality. As a result of the increase of the problem space, computational cost becomes expensive and performance of the optimization algorithms decrease. To overcome this problem, a self-adaptive Artificial Bee Colony (ABC) algorithm called “Self-adaptive Search Equation-based Artificial Bee Colony” (SSEABC) is proposed in this paper. In SSEABC, the canonical ABC is modified with two strategies which are self-adaptive search equation determination and incremental population size. The first strategy determines the appropriate search equations for a given problem instance adaptively during execution. On the other hand, incremental population size strategy adds new food sources to the population biased towards the best-so-far solution. This leads to performance improvement. SSEABC was tested on the benchmark set provided for the special issue of Soft Computing Journal on “Scalability of evolutionary algorithms and other metaheuristics for large-scale continuous optimization problems” (SOCO). We compared the results of the proposed algorithm to the five ABC variants and the fifteen SOCO participant algorithms. The comparison results indicate that SSEABC is more effective than the considered ABC variants and very competitive with regards to the fifteen SOCO competitor algorithms.

In recent years, complex-, quaternion-, and Clifford-valued neural networks have been intensively studied. This paper introduces Lie algebra-valued bidirectional associative memories, an alternative generalization of the real-valued neural networks, for which the states, outputs, and thresholds are all from a Lie algebra. The definition of these networks is given, together with an expression for an energy function, that is indeed proven to be an energy function for the proposed network.

The focus in this paper is on the proposal of guaranteed patterns in the training set for associative networks. All proposed patterns are pseudoortogonal and they also fulfil stability condition. Patterns were stored into the matrix using Hebb rules for associative networks. In the experimental study, we tested which from the heteroassociative Bidirectional Associative Memory (BAM) and autoassociative Hopfield network is more effective when working with the proposed patterns and what are the possibilities for Hopfield networks when working with real patterns. The comparison was made in order to recognize various damaged images using both types of associative networks. All obtained results are presented in tables or in graphs.

In this paper, the latest results of research in the area of author’s personal style profile construction are reviewed. The main goal is to explore the ability to use Markov chain graph, educated based on original author texts to store specifics of his personal writing features. Having such personal profile enables text comparison for authorship confirmation. The ability to do it will be in demand in lot of different areas, for example, authorship detection of scientific articles, or artistic literature texts. This paper describes the main idea offered, the proposed algorithm for two graphs similarity level calculation, the structure of the experimental system created and results of the experiments conducted.

Dust storm phenomena have vital effects on human life and are significant threat on ecosystem, climate and environmental conditions. Therefore, it may be of vital importance to develop an effective prediction system and mechanism to prevent it and/or reduce its devastating effects. This paper focuses on predicting meteorological conditions associated with dust-storms in the city of Ahvaz, south-western of Iran utilizing local linear neuro fuzzy model with LOLIMOT learning algorithm. For this purpose two different cases are considered. The first case aims to predict the next storm day occurrence and the second case focuses to calculate the number of storm days in next 15 days. The results show that findings under both cases are very close to reality and efficient for predicting dust storm occurrences in Ahvaz city and thus, the methodology could be useful for predicting this event for similar cities as well.

This paper considers the problem of solving the Maximum Traveling Salesman Problem (otherwise known as “taxicab ripoff problem”) in a plane, using an adapted Neural Gas algorithm with some features of Kohonen’s Self-Organizing Map. Maximum Traveling Salesman Problem is similar to classical Traveling Salesmen Problem (TSP), but instead of a search for a Hamiltonian tour visiting all vertices and returning to the initial vertex, which has a minimum sum of lengths of visited edges, we are looking for a maximum sum of lengths. This problem is less popular than classical TSP, nevertheless, in recent years also received a great amount of attention, leading to many heuristics and theoretical results. In this paper, we propose a new heuristic, which is certainly not as efficient as the already existing methods. We are not trying to enter the fierce competition to find the most effective algorithm, but we are trying to experimentally examine a possibility to use a special type of neural network to solve such a problem. Experiments show, that elements of neural gas approach together with SOM are applicable to this kind of problem and provide reasonable results.

This paper introduces conjugate gradient algorithms for training quaternion-valued feedforward neural networks. Because these algorithms had better performance than the gradient descent algorithm in the real- and complex-valued cases, the extension to the quaternion-valued case was a natural idea. The classical variants of the conjugate gradient algorithm are deduced starting from their real-valued variants, and using the framework of the HR calculus. The resulting quaternion-valued training methods are exemplified on time series prediction applications, showing a significant improvement over the quaternion gradient descent algorithm.

This paper proposes a fuzzy system for selection of suitable job applicants using an expert system. We propose the evaluation of job applicants using evaluation number and the evaluation of a job interview by a prepared questionnaire and expert system – EXS1. Based on this information and knowledge base, the expert system (EXS2) suggests the most suitable candidates for the position. The proposed fuzzy system is verified on a selected job position and a few job applicants.

Temporal association patterns are those patterns which are obtained from time stamped temporal databases. A temporal association pattern is said to be similar if it satisfies specified subset constraints. The apriori algorithm which is designed for static databases cannot be extended to find similar temporal patterns from temporal databases as patterns are vectors with supports computed at different time slots and Euclidean distance do not satisfy monotonicity property. The brute force approach to find similar temporal patterns requires computing \(2^n\) true support combinations for ‘n’ items from finite item set and problem falls in NP-class. In this present research, we come up with novel approach to discover temporal association which are similar for pre-specified subset constraints, and substantially reduce support computations. The proposed approach eliminates computational overhead in finding similar temporal patterns. The results prove that the proposed method outperforms brute force approach.

Mining Temporal Patterns from temporal databases is challenging as it requires handling efficient database scan. A pattern is temporally similar when it satisfies subset constraints. The naive and apriori algorithm designed for non-temporal databases cannot be extended to find similar temporal patterns from temporal databases. The brute force approach requires computing \(2^n\) true support combinations for ‘n’ items from finite item set and falls in NP-class. The apriori or fp-tree based approaches are not directly extendable to temporal databases to obtain similar temporal patterns. In this present research, we come up with novel approach to discover temporal association patterns which are similar for pre-specified subset constraints, and substantially reduce support computations by defining expressions to estimate support bounds. The proposed approach eliminates computational overhead in finding similar temporal patterns. The results prove that the proposed method outperforms brute force approach.

Missing values are quite common in medical records. Fixing missing values is a challenging task to data mining analysts as any error in imputation process leads to potential hazards. In the present research, the main objective is to impute missing values using a new similarity measure and applying class based cluster approach which is used to perform dimensionality reduction. The proposed approach is demonstrated using an effective case study. The results show the proposed measure performs better and is efficient.

This paper deals with mobile robot localization purpose. The presented solution is designed for indoor environment only. GPS navigation system cannot be used in environment inside of buildings. Alternative methods have to be used for this purpose. The mobile robot localization is essential part of autonomous mobile robotics. Mobile robot localization together with odometry is necessary for mobile robot navigation. Presented paper contains explanation of localization approach, which is based on probabilistic method. Next part of this paper is description of experimental odometry method, which is based on computer vision.

This article presents a hybrid method for a processing of a cloud point. Proposed method is suitable for reverse engineering where the need of precise model representation is essential. Our method is composed of mathematical representation using T-spline surfaces and edge extraction using k-neighborhood and Gauss mapping. The advantages of this method that we are able to find mathematical expression of the model where modification of parameters expresses the edges directly.

This paper deals with an estimating of the Fractal Dimension of a hydrometeorology variables like an Air temperature or humidity at a different sites in a landscape (and will be further evaluated from the land use point of view). Three algorithms and methods of an estimation of the Fractal Dimension of a hydrometeorology time series were developed. The first results indicate that developed methods are usable for the analysis of a hydrometeorology variables and for a testing of the relation with autoregulation functions of ecosystem.

The paper continues in previous works published by the authors where the emergent situations were detected by the violence of structural invariants. In this paper is used only one type of structural invariant – Matroid and Matroid Bases (M, BM) investigating the influence of their violation to interacting elements (components) in so called basic group (compartment). The application of the presented approach is demonstrated in cases of discovery of diseases (diseases of cardio-vascular system). In the second plan of this paper is introduced a new method of the discovery of a semantic content of emergent shapes in ECG signals. The method is illustrated in the case of cardiac arrhythmia diagnosis.