Table of Contents

Theses related to FAKE GAME

Here is a list of theses related to FAKE GAME.

FAKE GAME

P. Kordík: Fully Automated Knowledge Extraction using Group of Adaptive Models Evolution

Pavel Kordík: Fully Automated Knowledge Extraction using Group of Adaptive Models Evolution (Doctor thesis, english)

Keywords like data mining (DM) and knowledge discovery (KD) appear in several thousands of articles in recent time. Such popularity is driven mainly by demand of private companies. They need to analyze their data effectively to get some new useful knowledge that can be capitalized. This process is called knowledge discovery and data mining is a crucial part of it. Although several methods and algorithms for data mining has been developed, there is still a lot of gaps to fill. The problem is that real world data are so diverse that no universal algorithm has been developed to mine all data effectively. Also stages of the knowledge discovery process need the full time assistance of an expert on data preprocessing, data mining and the knowledge extraction.

These problems can be solved by a KD environment capable of automatic data preprocessing, generating regressive, predictive models and classifiers, automatic identification of interesting relationships in data (even in complex and high-dimensional ones) and presenting discovered knowledge in a comprehensible form. In order to develop such environment, this thesis focuses on the research of methods in the areas of data preprocessing, data mining and information visualization.

The Group of Adaptive Models Evolution (GAME) is data mining engine able to adapt itself and perform optimally on big (but still limited) group of real-world data sets. The Fully Automated Knowledge Extraction using GAME (FAKE GAME) framework is proposed to automate the KD process and to eliminate the need for the assistance of data mining expert. The GAME engine is the only GMDH type algorithm capable of solving very complex problems (as demonstrated on the Spiral data benchmarking problem). It can handle irrelevant inputs, short and noisy data samples. It uses an evolutionary algorithm to find optimal topology of models. Ensemble techniques are employed to estimate quality and credibility of GAME models.

Within the FAKE framework we designed and implemented several modules for data preprocessing, knowledge extraction and for visual knowledge discovery.

Preprocessing

M. Pavlíček: Automated Data Preprocessing Module (czech)

Miloslav Pavlíček: Modul pro automatické předzpracování dat (Master thesis, czech)

Data preprocessing is essential step in data mining process, however is extremely time consuming. This diploma thesis studies ways of automatic selection of preprocessing methods, so the results of data mining process are best possible.

Outcome of this work is a module for application FAKE GAME, which is used for developing of sequences of data preprocessing methods and setting their parameters. It is possible to apply those sequences to each column of input data separately. Module operates as pre-stage of GAME models, which determine its success by error rate on input data. Module also allows to export best sequences for later use, particularly for genetic programming.

H. Krkošková: Data Reduction for Data Mining and the GAME Neural Network (czech)

Helena Krkošková: Redukce dat pro data mining a neuronová síť GAME (Master thesis, czech)

Training of the GAME neural network is very time consuming task. Time required to create a successful model increases with size of training set. This thesis investigates methods for data reduction (data condensing). Presented methods reduce number of instances in training set and in this way methods reduce time required to train GAME neural network and also reduces amount of memory required during the training process. The important requirement set on condensing methods is that reduced training set must produce a model with the same or similar accuracy as a model created on the not reduced set.

Result of this work is implementation of selected methods for data condensing into data preprocessing module of the FAKE GAME project. The implemented condensing methods were tested using the GAME neural network and 1-NN classifier on several real-world and artificial datasets.

D. Meško: Data Normalization for the GAME Neural Network (czech)

Dezider Meško: Normalizace dat pro neuronovou síť GAME (Bachelor thesis, czech)

This thesis deals with the linear and nonlinear input data normalization before it is used for neural networks calculation. These normalizations are implemented into FAKE GAME system and the implementation is afterward tested. Its influence on GAME models is generally compared through several testing data files.

Optimization

O. Kovářík: Ant Colony Optimization for Continuous Problems

Oleg Kovářík: Ant Colony Optimization for Continuous Problems (Master thesis, english)

This work includes an overview and tests of optimization algorithms in the continuous domain. Algorithms are inspired by the ant behavior, mainly by coordination of their work by using pheromone layers. Algorithms covered in the thesis are: Ant Colony Optimization (ACO), Ant System (AS), Continuous Ant Colony Optimization (CACO), API, Continuous Interacting Ant Colony (CIAC), Adaptive Ant Colony Algorithm (AACA, BACA), Binary Ant System (BAS), Extended Ant Colony Optimization (ACO*), Continuous Ant Colony System (CACS), Aggregation Pheromone System (APS), Direct ACO (DACO). Methods with similar approach are grouped and representatives of the groups are implemented, evaluated on the real world data and compared with other continuous optimization methods. The most successful out of the implemented ant behavior inspired methods were DACO and ACO* that had very good results, although they didn’t go better then Quasi-Newton method.

M. Hvizdoš: Continuous Optimization Library (slovak)

Martin Hvizdoš: Knihovna pro spojitou optimalizaci (Master thesis, slovak)

Optimization problems are ubiquitous. From data analysis and financial planning to modeling of chemical processes or design. There are many ways to their solution which range from purely numerical, hybrid to those inspired by nature. The diversity and complexity of the individual approaches and implementations is causing problems in understanding them. This thesis focuses on creations of instruments for optimization method testing and inspection. One of its main goals is simplicity. Simplicity not only for the end user but also for the optimization problem and method implementation creator. And thus seeks to attract both.

O. Filípek: Conjugate Gradient Method for GAME Units Optimization (czech)

Ondřej Filípek: Metoda sdružených gradientů pro optimalizaci jednotek GAME (Master thesis, czech)

During construction of a model from data is necessary to set parameters of the model so it will closely describe behavior of the system which the data describes. There are many methods to solve this, but it’s very difficult to choose the right one which would be the most effective. In this work we try to use conjugate gradient method. Than we are looking for the best configuration for it. The optimization module based on conjugate gradient method has been implemented to the GAME. It also allowed to compare performance of this module with already implemented modules.

M. Janošík: GAME Networks Optimization Algorithms (czech)

Miroslav Janošík: Algoritmy pro optimalizaci sítí GAME (Bachelor thesis, czech)

This bachelor thesis is about a problem of optimization. Concretely it is about optimization of units of neural network GAME. This thesis includes theoretical part, where is the description of some optimization methods, and it also includes practical part. There was created an implementation of algorithm of differential evolution. This implementation was tested on some chosen data sets. On the same data sets were applied other optimization methods too. This results was compared to results of algorithm of differential evolution. High quality of created algorithm was proved.

Modeling

J. Špirk: Distributed Computation of Inductive Models (czech)

Jakub Špirk: Distribuovaný výpočet induktivních modelů (Bachelor thesis, czech)

The goal of this thesis is to introduce inductive modeling, especially GAME method, as one of the methods used in data mining process and compare this method with other methods of data mining. In the next part I would like to present terms like parallelism and parallel programing on Symmetric multiprocessing systems, explain the difference in basic approaches of finding parallel solutions, discuss the possibilities and assets of parallelizing evolution of inductive models generated by GAME method and implement the chosen solution in JAVA programming language. The goal is not to implement new natural evolution algorithms, the ones used in GAME method are to be preserved.

D. D. Minh: Cascade Correlation Neural Networks (czech)

Minh Duc Do: Neuronové sítě Cascade Correlation (Master thesis, czech)

The goal of this work is to implement a general library in language Java for neural network, which would be using the learning algorithm Cascade-Correlation. Cascade-Correlation itself uses other learning algorithms and during the learning process it changes the structure of the neural network by adding new hidden layers. This learning algorithm is supposed to work faster in compare to the widely known Backpropagation. It should also provide better results in the case of more difficult problems.

At the end the library will be integrated into the project GAME for data mining as one of an alternative learning modul.

Visualization

M. Chren: Automatic Knowledge Extraction (slovak)

Michal Chren: Automatická extrakcia znalostí (Master thesis, slovak)

This thesis focuses on three consequential topics. The first one is a visualization of 3D slices of GAME models and training data. The models are defined as functions of n inputs and the data are points in this (n+1)-dimensional space. The 3D slice is defined by the selection of two of the inputs as its axes and by setting the remaining n – 2 inputs to constant values.

The second problem is the search for such slices that are of some interest to the users. This means that graphs representing individual models have highly varying values on their domain of definition. Moreover, since we are talking about a 3D slice out of an n – 1-dimension space, the search area is very large. Therefore, I used a special type of genetic algorithm, known as Deterministic Crowding. Its advantage over the classic genetic algorithm is that it is able to find and keep several different and suboptimal solutions. Automatic knowledge extraction is realized in this way – in fact, the models represent data abstraction and the algorithm finds 3D views, in which the model outputs are similar (the abstraction is reliable) and, at the same time, their shapes are complicated.

The third theme covered by this thesis is report generation and automatic reporting. The reports should contain descriptions of slices found in the above-mentioned search that are of any interest and their visualization, thus mediating an intuitive and practical overview of the trained models as well as of the knowledge extracted from the data. The conclusion of this thesis is a module in the FAKE GAME application, written in Java language, which successfully implements all the mentioned issues.

J. Matejička: Statistical Properties of Data in GAME

Jozef Matejička: Statistical properties of data in GAME (Master thesis, english)

In recent years there is huge interest in data mining software from both industry and science. Fake Game is data mining software developed on Faculty of Electrical Engineering, Czech technical university. One of biggest problem in the field is to overcome curse of multi-dimensionality. One of the approaches to overcome this is to use expertise of user and decrease number of dimensions in input data.

This work focuses on implemented numerous well established visualization techniques. Visualizations should help user of Fake game to expertly decide if it is desirable to remove some data. Other visualization tools compare models according their quality.

J. Matoušek: Visualizing Topology of Hybrid Inductive Models (czech)

Jaroslav Matoušek: Zobrazení topologie hybridních induktivních modelů (Master thesis, czech)

GAME engine generates models capable to analyze even very complex data automatically. By studying the topology of created models it is possible to get new knowledge about the data character. To examine the quality of models it is useful to visualize their responses in a wellarranged way.

The outcome of this thesis is a visualization module for FAKE GAME application enabling 3D display of inductive models. To check the quality of the model the visualization of response to input data vectors was implemented. The response can be viewed either for each unit separately or for all units at once in a topology order. Extension of the 2D visualization by adding SVG export is also part of the thesis.

J. Nožka: Visual 3D Analysis of Neural Networks Behavior (czech)

Jiří Nožka: Vizuální 3D analýza chování neuronových sítí (Master thesis, czech)

The GAME method automatically generates models of variously complex systems. Models built by this method are often multidimensional, just as the data, which are used in training these systems. The problem is to find out, whether the output of model approximates the training data correctly.

The aim of this work is to get better insight into a multidimensional behavior of models, by means of 3D graphics. As the multidimensional space is too extensive, we applied genetic algorithms to locate remarkable areas in this space automatically.

M. Škola: Visualizing Data for Knowledge Extraction (czech)

Michal Škola: Vizualizace dat pro exktrakci znalostí (Master thesis, czech)

Thesis deals with non-trivial knowledge extraction using visualizations by data mining. With JFreeChart library implements charts to FAKE GAME and allows export them to the vector graphics. Describe knowledge, which could we extract from plots.

J. Saidl: Visualization as a Tool for Studying Behavior of Natural Systems Models (czech)

Jan Saidl: Vizualizace jako nástroj studia chování modelů přírodních systémů (Master thesis, czech)

The common topic of all experiments in this thesis is the visualization of neural network behavior. We had some difficulties dealing with the multidimensional character of the models, therefore we had to study and apply several concepts for multivariate data projection. Approaches proposed in this thesis are unique, so we spend a lot of time tuning them to be applicable to real-world problems. Finally we applied genetic algorithm preserving diversity to find interesting behavior of neural network in multidimensional space.

Evaluation

J. Kuchař: Statistical Data Processing Module (czech)

Jaroslav Kuchař: Modul pro statistické zpracování dat (Master thesis, czech)

This thesis designs and implements statistical data processing module in Java. This module provides statistical evaluating of inductive models. Part of this thesis is introduction to statistical interface of the R language. R is measurement standard in statistical world. Statistical evaluating can evaluate correctness of the model learning. The first part is retrieval of the statistical libraries, models theory and statistical theory. Second part is aimed at implementation and integration statistical methods into module.

Import / Export

V. Uličný: Hybrid Inductive Models in PMML

Vít Uličný: Hybrid inductive models in PMML (Master thesis, english)

Data mining became very popular, many algorithms for creating data mining and predictive models exist, but problem is how to share these models. Standard PMML exists, which is defined for some common models. No support exists for sharing inductive models with hybrid units. This thesis deals with an extension of standard and implementation of a support to the application GAME for serialization to this data format.

D. Dvořák: Application for Visualizing Structure and Behavior of Inductive Models (czech)

David Dvořák: Aplikace pro vizualizaci struktury a chování induktivních model (Master thesis, czech)

Data mining has become very popular, there are many algorithms for data mining models creation, the problem is sharing these models. There is a PMML standard, which is defined for some models. But no support for a hybrid inductive models exists. This thesis deals with an extension of PMML standard and implementation of necessary functions for visualization of structure and behavior of loaded inductive models.

P. Král: Importing and Exporting Data from/to MS Excel and Openoffice Spredsheets (slovak)

Peter Král: Import a export dát z/do MS Excel a Openoffice Spreadsheet

The function of this work is to explore the spreadsheet file formats (XLS, ODS). Then use these experiences for the implementation of import/export spreadsheet data in to the GAME system. The first part of this work is aimed at the explanation of the spreadsheet file formats and the second part is aimed at the design of the solution and its implementation.

Miscellaneous theses related to FAKE GAME

J. Weberschinke: Design Patterns in the FAKE GAME Application (czech)

Jakub Weberschinke: Návrhové vzory v aplikaci FAKE GAME (Bachelor thesis, czech)

This thesis contains introduction to design pattern issues with focus on GoF and GRASP patterns. The next part contains description of basic terms of data mining with characterization of Fake Game and Weka applications. Describes the most used methods of evaluation of data mining models and compares the former applications in this respect. Introduces current problems of Fake Game API and discusses possible solutions. The last part is a description of an integration of Weka evaluation module into Fake Game.

V. Šťastný: Web Services in the FAKE GAME Application (czech)

Vladimír Šťastný: Webové služby aplikace FAKE GAME (Bachelor thesis, czech)

At first, this bachelor thesis deals with the web services in the complex view. It focuses on the main principles which are essential for the web services. Also, the existing concept of web services based on the grid technology is described. Then, the thesis is concerned with the FAKE GAME application. The main effort is to introduce the FAKE GAME application so that it could be used in the practical part of the thesis. At the end, a possible concept of web services relating to the FAKE GAME application is described.

M. Trčka: Simplification of Formulas Generated by Inductive Models (czech)

J. Bouška: Neural Networks for Time Series Prediction (czech)

J. Bouška: Neuronové sítě pro predikci časových řad (Master thesis, czech)

The goal of this diploma thesis is to find the most appropriate method from computational intelligence domain for slow biological signals prediction. To achieve this goal we need to study the basics of time series analysis, prediction by means of inductive models, choose the proper representatives from parametric and nonparametric methods and perform experiments by means of selected methods on various types of real data sets, including biological data from Motol Hospital in Prague. The results from this work can contribute to treatment of patients with brain injury.

M. Čepek: Pattern Recognition for Real World Data

Miroslav Čepek: Pattern Recognition for Real World Data (Master thesis, english)

A long time sleep EEG recording is very important in diagnosis of many diseases and sleep difficulties. For easier diagnosis it is necessary to know so called sleep stages. Because manual Sleep stages scoring is a time consuming work, there is a constant desire to automate the process. The automation is discussed in many works, but the most of them uses only Fourier transform to extract relevant features from the signal. In this work I will try to find out if some new features do not describe the Sleep stages better. Among many feature extraction methods the Wavelet transform, Statistic and few others will be used. After the feature extraction, useful features will be selected with various data mining methods and then each selection will be evaluated with several classifiers.

M. Hasaj: Smart Home − Teiresias

Martin Hasaj: Smart Home -- Teiresias (Master thesis, english)

A great deal of contemporary research is showing that it is not work that goes home but home that goes to work. I would like to write my thesis about smart home possibilities (theoretically), which are available, or proposed to the market. Then I would like to design (practically) part of smart home according to available technologies. I am more considered with the software part of the problem, but I want to use real hardware if it would be possible rather then a simulator.

Software should be portable, written in JAVA, using Open Source Solutions and Technologies. It should be able to manage all smart home interfaces and extensible. Brain of the system should be Neural Network, which should be able to learn automatically the inhabitants' behaviors and help them in their everyday routine. “Make technologies disappear and you have succeeded.”

For example, if system finds out that some switch is turning on every time after doors are open, it will turn it on automatically. But it should consult this decision in the beginning to avoid collisions with human decision (according to Asimov rules). These shall avoid starting mixer with nothing in it, or even turning lights on when nobody is at home.

The system shall communicate through the network, which could be wired, or wireless. After searching the technologies available I decided to concentrate on open standard KNX.

J. Novák: GMDH Networks the KnowledgeMiner Software

Jakub Novák: GMDH networks the KnowledgeMiner software (Bachelor thesis, english)

Inductive methods such as Group Method of Data Handling (GMDH) are good in solving ill-posed tasks. Our past studies indicate that the task of the age estimation based on senescence indicators obtained from skeletons is ill-posed. In this work I apply the GMDH methods implemented in the KnowledgeMiner software on the “Antro” and “Building” data. I also compare the KnowledgeMiner to another application the GAME simulator.

J. Novák: Analyzing Anthropological Data using Computational Intelligence Methods (czech)

Jakub Novák: Analýza antropologických dat metodami výpočetní inteligence (Master thesis, czech)

A computational intelligence methods are suitable instrument for work with the anthropological data which represents senescence indicators along with other inputs. Based on this information we try to predict the age of skeleton. But this is a very difficult process and obtain high-quality results is complicated. My goal in this diploma thesis is to find and valorize the best methods which can handle well the anthropological data and give us the best results.

A. Pilný: Processing Biological Signals (czech)

Aleš Pilný: Zpracování biologických signálů (Master thesis, czech)

A new examination method of patients' balance dysfunctions detection is being developed in the Motol Faculty Hospital, Prague. During this examination, patients are sitting on a special rotating chair. Their eye movements together with the driving signal of the chair are recorded. For healthy patients, the eye movement signal should contain the same frequencies as the driving signal does. However, the eye movement signal contains a movement correction, called saccades, which make the direct application of the FFT for obtaining the frequencies unsuccessful. In this work, we present a new method of saccade detection using computational intelligence (algorithm GAME). We have also implemented a new program RecBob for saccade removal and signal reconstruction. The proposed method is experimentally validated on real data from Motol hospital. All frequencies contained in the signal were successfully identified.

P. Samek: Evaluating Results of a Comparative Study in a Recondition Center (czech)

Petr Samek: Vyhodnocení výsledků srovnávací studie v rekondičním centru (Master thesis, czech)

The aim of this diploma thesis is to analyze the data collected from pilot study “reconditioning of patients suffering from serious obesity” run 1st Medical Faculty of the Charles University in Prague. The results of the thesis discuss influence of measured variables to fat reduction and other parameters. In the second part of the thesis we try to identify the most accurate formula predicting parameters of interest based on variables measured during the initial examination of the patient. Data are analyzed not only by statistical methods but also by means of methods from the computational intelligence domain and results are interpreted for the expert.

L. Tůma: Analyzing Signals of Biological Neurons (czech)

Lukáš Tůma: Analýza signálů biologických neuronů (Master thesis, czech)

Single Units are electrical phenomena recorded in the vicinity of excitable cells (biological neurons) or inside them. Evaluation of single units waveform is burdened with enormously time-consuming labour. In this work an algorithm of their semi-interactive tracking, sorting and comparing of entire recordings among each other was created and implemented. Tracking of single units is conducted using characteristic features (amplitude, steepness of amplitude increase, asymmetry of elements) of points of the record. These can be adjusted to relevant needs. Selected points are further compared and sorted on a modified artificial neuronal network (K-means, network without teacher) with repeated possibility of selection of desired subpopulation. In the next step, we compare and sort compare and sort individual entire recordings among each other according to incidence of selected elements, their frequency median, amplitude and frequency histogram. The program implements neither sole proceeding stated above but also a table output facilitating further computations (e.g. statistics) of outputs of particular sections of the program.

 
theses.txt · Last modified: 2009/10/06 07:33 by tregoreg
 
Except where otherwise noted, content on this wiki is licensed under the following license:CC Attribution-Noncommercial-Share Alike 3.0 Unported
Recent changes RSS feed Donate Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki