Showing papers by "Milan Tuba published in 2012"

Artificial Bee Colony (ABC) Algorithm for Constrained Optimization Improved with Genetic Operators

Abstract: Artificial bee colony (ABC) is a relatively new swarm intelligence based metaheuristic. It was successfully applied to unconstrained optimization problems and later it was adjusted for constrained problems as well. In this paper we introduce modifications to the ABC algorithm for constrained optimization problems that improve performance of the algorithm. Modifications are based on genetic algorithm (GA) operators and are applied to the creation of new candidate solutions. We implemented our modified algorithm and tested it on 13 standard benchmark functions. The results were compared to the results of the latest (2011) Karaboga and Akay’s ABC algorithm and other state-of-the-art algorithms where our modified algorithm showed improved performance considering best solutions and even more considering

Performance of a Modified Cuckoo Search Algorithm for Unconstrained Optimization Problems

Abstract: Cuckoo search (CS) algorithm is one of the latest additions to the group of nature inspired optimization heuristics. It has been introduced by Young and Deb in 2009 and was proven to be a promising tool for solving hard optimization problems. This paper presents a modified cuckoo search algorithm for unconstrained optimization problems. We implemented a modification where the step size is determined from the sorted, rather than only permuted fitness matrix. Our modified algorithm was tested on ten standard benchmark functions. Experimental results show that our modification improves results in most cases, compared to the pure cuckoo search algorithm.

Adjusted artificial bee colony (ABC) algorithm for engineering problems

Abstract: In this paper we present a modified algorithm which integrates artificial bee colony (ABC) algorithm with adaptive guidance adjusted for constrained engineering optimization problems. The novel algorithm improves best found solutions in some cases and improves robustness i.e. mean value and variance for number of runs in other cases by improving the algorithm's exploitation/exploration balance. Even though scout bee phase is used for exploration, we introduced adaptive parameter that at different stages of the algorithm narrows search space facilitating faster convergence. We tested our algorithm on four standard engineering benchmark problems. The experimental results show that our modified algorithm can outperform the pure ABC algorithm in most cases.

Parallelized cuckoo search algorithm for unconstrained optimization

Abstract: Modifications that introduce parallelization of standard cuckoo search algorithm are proposed in this paper. Basic form of the cuckoo search algorithm has already shown great potential for optimization problems, especially when applied to unconstrained continuous functions. In this paper two aspects of parallelization are proposed. The first one addresses the performance issue, while the second one deals with quality of results. Multicore processors became standard today. When different runs of algorithm execute within different threads, better performance can be reached. Second issue refers to multiple flocks approach that combines search results from two or more flocks. Set of well-known unconstrained continuous benchmark function is used to illustrate testing results of the proposed parallelized cuckoo search algorithm.

Swarm intelligence algorithms parameter tuning

Abstract: Nature inspired metaheuristic algorithms are recently successfully used to find suboptimal solutions to hard optimization problems. These algorithms mimic different nature phenomena in hope that nature's implicit intelligence will help to guide the search in untractable problems. Swarm intelligence algorithms are a class of nature inspired algorithms based on collective intelligence of colonies of ants, bees, fish etc. They have a number quantitative and qualitative parameters that can be adjusted. Such adjustments are not allowed for specific test problems but only for a whole class. When some adjustment works for number of classes it can be incorporated into the generic algorithm as a new qualitative parameter (optional modification). In this paper we describe how successful application of pheromone correction strategy for the ant colony optimization (ACO) algorithm on three different graph problems is incorporated in ACO software framework as a module.

Parallelization of the local threshold and Boolean function based edge detection algorithm using CUDA

Abstract: In this paper we present a parallelized algorithm for edge detection for gray scale images. The chosen method is the local threshold and boolean function based edge detection. This method differs from common edge detectors in the use of bit map patterns instead of analyzing gradient changes in the image for edge recognition. The parallelization is implemented on the GPU, exploiting its multithreaded, many-core processor power using NVIDIA's CUDA (Compute Unified Device Architecture). We show in our tests the significant speedup of parallelized algorithm compared to the sequential one.

An object-oriented implementation of the firefly algorithm for continuous unconstrained optimization problems

Abstract: Firefly algorithm is one of the latest additions to the family of swarm intelligence metaheuristics for hard optimization problems. It is based on the mating and social behavior of fireflies. It is not well researched yet and its potentials are undetermined. This paper describes an object-oriented implementation of the firefly algorithm for continuous unconstrained optimization problems. It was tested on nine standard benchmark function and results are promising compared to other swarm intelligence algorithm. This software framework will allow for easy modifications and further testing with the possibility that this algorithm will outperform older algorithms.

Modified seeker optimization algorithm for unconstrained optimization problems

Abstract: Seeker optimization algorithm (SOA) is a novel search algorithm based on simulating the act of human searching, which has been shown to be a promising candidate among search algorithms for unconstrained function optimization In this article we propose a modified seeker optimization algorithm In order to enhance the performance of SOA, our proposed approach uses two search equations for producing new population and employs modified inter-subpopulation learning phase of algorithm This modified algorithm has been implemented and tested on fourteen multimodal benchmark functions and proved to be better on majority of tested problems

Plenary lecture 3: swarm intelligence algorithms parameter tuning

Abstract: Most real-life problems can be represented as some kind of optimization problem. Easy optimization problems were solved long time ago so nowadays only hard problems are of research interest. Many discrete (combinatorial) as well as some continuous optimization problems are intractable, but of great practical interest. Traveling salesman problem (TSP) is a classic example that was researched for the longest period of time and because of that is often used as a benchmark. The main problem with hard optimization problems is that there is enormous number of suboptimal solutions or local minima and there is no guidance how to search. Standard down-hill methods in this situation fail. Typical example of such function that is used as a benchmark is Rastrigin function that is a sphere modified by small cosine waves. The oldest way to deal with such problems is Monte-Carlo method. It is equivalent of trying to find the deepest point in the oceans by measuring many times the depth at random locations and hoping that best measurement will be close to the global optimum. While Monte-Carlo method is usable for some applications, its blind search is not sufficient for many others. In this rather hopeless situation researchers turned from mathematically exact methods to belief. The nature is doing miraculous things. We know the results but we do not understand the mechanism. For hard optimization problems we try to mimic some nature processes. Older attempts included simulation of evolution (through genetic modifications and survival of the fittest) and simulated annealing. Recently, swarm intelligence become prominent using the fact that extremely simple individuals exhibit miraculous collective intelligence. Examples include ants colonies, honey bees colonies, flocks of birds, schools of fish etc. These nature inspired metaheuristics simulate various natural phenomena. We talk about bee colony food finding or ant colony path finding, but in essence, in all these diverse mimicking we do two things. We exploit good found solutions, but also go to unknown places in order to avoid being trapped in local minima. The successfulness of any such algorithm is determined by proper balance between exploitation and exploration. This balance is maintained by adjusting certain parameters and also by applying some rules in certain situations. By doing such adjustments algorithm can become much better for some class of problems (off course, according to NFL theorem, it cannot become universally good for all problems). This plenary lecture will demonstrate few successful examples of such adjustments.

Artificial fish swarm algorithm for unconstrained optimization problems

Abstract: This paper presents an object-oriented software system that implements a modified artificial fish swarm (AFS) algorithm based on a particular intelligent behavior of schools of fish. We outline our implementation of the algorithm for unconstrained optimization problems. The application was implemented in C# with flexible GUI (Graphical User Interface) and it was successfully tested on four standard unconstrained benchmark problems.