|
|
Optimization of Multilayer Perceptron Training Parameters Using Artificial Bee Colony and Genetic Algorithm
Kartci, A.
In this paper, the momentum coefficient, learning rate, and the number of hidden neurons where the multilayer perceptron works best, are determined. The network and optimization algorithms are written in MATLAB, which was also successfully used to carry out results. To obtain the results, IRIS, mammographic_mass, and new_thyroid data sets have been used. Obtained results show that the determining effect on the neural learning process of parameters (momentum coefficient, learning rate, number of hidden neurons) are compatible with other approaches available in the literature. Both genetic algorithm (GA) and artificial bee colony (ABC) algorithm were successful on finding the values to get high performance as well as effect on performance of the population number.
|
|
|
Application of genetic algorithm for production scheduling of engineering company
Stariat, Jiří ; Skočdopolová, Veronika (advisor) ; Zouhar, Jan (referee)
This thesis is engaged in scheduling problem, his special types and methods of solving. Scheduling problem is a common operations research problem, which ranks among combinatorial problems. The aim of the scheduling problem is to assign certain activities and resources to individual time moments. Scheduling problem is NP-complete problem. Its computational complexity is thus so high, that there is currently no known algorithm that precisely solve its any instance in polynomial time. Is therefore used for its solution heuristics and metaheuristcs. In this thesis is described in detail metaheuristics of genetic algorithm. Application of genetic algorithm for production scheduling of specific engineering company is the main objective of this thesis.
|
|
|
A method for selecting a portfolio of tools for online marketing activities and supporting their management
Smutný, Zdeněk ; Doucek, Petr (advisor) ; Stříteský, Václav (referee) ; Novotný, Ota (referee) ; Hynek, Josef (referee)
Online marketing activities play an increasingly important role for organization in connection with the development of internet based technologies and their positive reception by the society. The aim of this dissertation is to design an artefact that would support the decision making of marketing specialists and thus the management of online marketing activities. The starting point is an explorative research among Czech companies, which identifies the issues felt as problematic and the needs of the selected set of organizations. Introduced at the same time is the current state of use of selected tools for online marketing by these organizations, and the situation is compared with worldwide development. The output of this explorative research, the examination of scientific literature, and a critical analysis serve as a basis for designing an own method, Genoma, whose purpose is to support the decision making of marketing specialists, and thereby also the management of marketing activities in internet-mediated environment. This method is presented as Deming (PDCA) cycle, which enables it to be used not only separately, but also as part of other frameworks for the management of marketing activities (e.g. the frameworks PMF, MCPF and RACE, which are presented in the dissertation). The Genoma method uses mainly the genetic algorithm for selecting a suitable portfolio of online marketing tools for a particular campaign. The selection is made on the basis of expected feedback at the level of social interaction, meeting the given marketing targets, and the financial demands of the individual tools. The prerequisite of using this method is a knowledge base that includes the area of sociotechnical interaction, which is based on interpreting phenomena related to the internet-mediated environment and the features of complex networks. Methodically, this dissertation builds on the complementary relationship of the behavioural (social informatics) and the design type of research (design science research). The final assessment of the suitability of the proposed method is done on the basis of a multiple case study, which uses also an own program created in C#, implementing the genetic algorithm used in the Genoma method.
|
|
|
Least squares method using genetic algorithm
Holec, Matúš ; Tichý, Vladimír (advisor) ; Šalamon, Tomáš (referee)
This thesis describes the design and implementation of genetic algorithm for approximation of non-linear mathematical functions using the least squares method. One objective of this work is to theoretically describe the basics of genetic algorithms. The second objective is to create a program that would be potentially used to approximate empirically measured data by the scientific institutions. Besides the theoretical description of the given subject, the text part of the work mainly deals with the design of the genetic algorithm and the whole application solving the given problem. Specific part of the assignment is that the developed application has to support approximation of points by various mathematical non-linear functions in several different intervals, and then it has to insure, that resulting functions are continuous throughout all the intervals. Described functionality is not offered by any available software.
|
|
|
Local approach in mechanical properties prediction
Brumek, J. ; Strnadel, B. ; Dlouhý, Ivo
Indentation technique was focused on the prediction of the strain hardening behaviour of carbide steels. An improved technique to determine the plastic properties of material from the load-displacement curve from a ball indentation test was proposed. The time severity for the search for an optimal solution for a non-linear constitutive model is dependent on a number of design variables. Common methods like gradient methods or linear programming can fail due the fact that they drop to the local minimum. The advantage of a genetic algorithm does not require knowledge of the target function. Proposed method was applied to the data from the instrumented indentation technique. Results were found to be in good agreement with the data from conventional, standard tests, and in less time.
|
|
|
Heuristic and metaheuristic methods for travelling salesman problem
Burdová, Jana ; Kalčevová, Jana (advisor) ; Zouhar, Jan (referee)
Minimal length of a travelling salesman's problem had been studied in this diploma these. Travelling salesman must come trough each place just once and then go back to the starting place. This problem can be illustrated as a problem of graph theory, such that places are the vertices, roads are the edges, distances of roads are the lengths of edges. The optimal travelling salesman's problem tour is the shortest Hamiltionian cycle in the graph. It is a classical NP-complete problem. There is no algorithm that solves this problem in polynomial time. This problem can be solved by using various approximation algorithms, they offer less time consumption and lowest quality than optimization. This diploma these had been dedicated to approximation algorithms, for example: nearest neighbor method, minimal spanning tree method, Christofide's method, 2-opt., genetic algorithm, etc.
|
|
| |
|
| |
|
| |
|
|
Genetická selekce a klonování u metody GMDH-MIA
Jiřina, Marcel ; Jiřina jr., M.
The GMDH MIA algorithm is modified by the use of selection procedure from genetic algorithms and including cloning of the best neurons generated. The selection procedure finds parents for a new neuron among already existing neurons according to fitness and with some probability also from network inputs. The essence of cloning is slight modification of parameters of copies of the best neuron, i.e. neuron with the largest fitness. The genetically modified GMDH network with cloning (GMC-GMDH) can outperform other powerful methods. It is demonstrated on some tasks from Machine Learning Repository.
|
guest ::
