|
|
Application of the Artificial Intelligence in the Real Estate Valuation
Štechová, Edita ; Witzany, Jiří (advisor) ; Fičura, Milan (referee)
The main purpose of this study is to develop a predictive model capable to forecast residential real estate prices in the city of Prague using Artificial Intelligence methods. The first part of this study discusses fundamentals of Artificial Neural Networks and Fuzzy Inference Systems in the context of real estate valuation. The second part demonstrates a development and testing of such models using a dataset of real estate market transactions. In the third part, results are compared to Multiple Regression and an explanatory power of each model is evaluated. Conclusions of this research are: (1) Artificial Neural Networks and Fuzzy Inference Systems give more accurate estimates of market values of residential real estates than Multiple Regression; (2) Artificial Neural Networks and Fuzzy Inference Systems represent an efficient way of modeling and analyzing residential real estate prices in Prague.
|
|
|
Analysis of selected indicators on stock market
BUREŠ, Otto
In this work was evaluated the effectiveness of artificial neural networks in trading on the stock markets. The subject of the work was the process of optimizing parameters of artificial neural networks, the resulting predictive efficiency was determined on the basis of the application being optimized parameters of neural networks.
|
|
|
Capabilities of Radial and Kernel Networks
Kůrková, Věra
Originally, artificial neural networks were built from biologically inspired units called perceptrons. Later, other types of units became popular in neurocomputing due to their good mathematical properties. Among them, radial-basis-function (RBF) units and kernel units became most popular. The talk will discuss advantages and limitations of networks with these two types of computational units. Higher flexibility in choice of free parameters in RBF will be compared with benefits of geometrical properties of kernel models allowing applications of maximal margin classification algorithms, modelling of generalization in learning from data in terms of regularization, and characterization of optimal solutions of learning tasks. Critical influence of input dimension on behavior of these two types of networks will be described. General results will be illustrated by the paradigmatic examples of Gaussian kernel and radial networks.
|
|
|
Gold investment - IT support
ZAHOŘ, Zdeněk
This thesis is about the investigation of the prediction with help of artificial neural networks. It contrasts other papers considering similar topic. This work is focused on the data collected to predict future outcomes. The data specifically deals with gold capital. It is possible to buy gold via the internet in a form of gold bars. The dealers themselves state that the price of gold is derived via the global prize index of gold and from the current value converted to the Czech Koruna rate of exchange. The aim of this work is to consider price forecast correlating input data. To achieve the desired outcome a system for the data collection, processing, and visualization has been designed. A system dealing with price prediction has been developed to show the accuracy of the results obtained.
|
|
|
Geometrochemistry vs Soft Computing of Mendeleev's Brain
Gottvald, Aleš
The role of projective geometry in nature remains somewhat enigmatic for centuries. It is very strange indeed, as the projective geometry is the mother of all geometries with more restrictive symmetry groups, as clearly recognized yet by seminal insights of Felix Klein, Arthur Cayley, Paul Dirac and other eminent scientists. We usually imagine that Euclidean geometry is primary for the geometrization of our (nonrelativistic) spaces, and the Euclidean-Pythagorean metric is natural for measuring the distances in such a space. However, how to measure distances in spaces associated with statistical thermodynamics or quantum mechanics? We show that projective geometry and associated "geometrochemistry" is manifest in nature. In particular, it offers a novel soft-computing rationale for recovering basic structure of Mendeleev's periodic table of chemical elements, and elucidates some mysteries of brain information processing, including a new understanding of Artificial Neural Networks.
|
|
|
AE source location by neural networks independent on material scale changes
Chlada, Milan ; Převorovský, Zdeněk
The localization of acoustic emission (AE) sources by procedures using artificial neural networks (ANN) represents today highly effective alternative approach to classical triangulation algorithms. The main problems are in the collecting sufficiently extensive training and testing data sets together with the non portability of particular trained network to any other object. Recently, the ANN based AE source location method has been improved by using so-called signal arrival time profiles to overcome both limitations. This way of signal arrival time characterization enables ANN training on numerical models and allows the application of learned ANN on real structures of various scales and materials. In this paper, the method is upgraded and localization results are illustrated on experimental data obtained during pen-tests on a model roof I-beam and an aircraft structure part. General application possibilities of the method variations for different sensor configurations are also discussed.
|
|
| |
|
|
Učení vícevrstvých perceptronů s po částech lineárními aktivačními funkcemi
Kozub, P. ; Holeňa, Martin
This paper presents an overview of the techniques used to solve constrained optimization problems using evolutionary algorithms. The construction of the fitness function together with the handling of feasible and infeasible individuals is discussed. Approaches using penalty functions, special representations, repair algorithms, methods based on separation of objective and constraints and multiobjective techniques are mentioned.
|
|
|
OPTIMIZED NUMBER OF SIGNAL FEATURES FOR IDENTIFICATION OF AE SOURCES
Chlada, Milan ; Převorovský, Zdeněk
Artificial neural networks (ANN) are effective instruments for identification of AE sources. The proper selection of extracted data features is complicated task in general data recognition. Standard AE signal parameters are often redundant or not relevant in recognition problem. Modifications of standard AE signal features are proposed in this paper as to reduce data redundancy. Set of extracted AE parameters is optimized by factor analysis and sensitivity analysis of recognizing neural networks. This optimization is illustrated by recognition of AE sources arising during fatigue tests performed on aircraft structure parts. Optimized AE signal features cover enough information with minimized number of parameters.
|
|
|
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 ::
