National Repository of Grey Literature 8 records found  Search took 0.00 seconds. 
Transforming hierarchical images to program expressions using deep networks
Křen, Tomáš
We present a technique describing how to effectively train a neural network given an image to produce a formal description of the given image. The basic motivation of the proposed technique is an intention to design a new tool for automatic program synthesis capable of transforming sensory data (in our case static image, but generally a phenotype) to a formal code expression (i.e. syntactic tree of a program), such that the code (from evolutionary perspective a genotype) evaluates to a value that is similar to the input data, ideally identical. Our approach is partially based on our technique for generating program expressions in the context of typed functional genetic programming. We present promising results evaluating a simple image description language achieved with a deep network combining convolution encoder of images and recurrent decoder for generating program expressions in the sequential prefix notation and propose possible future applications.
Fulltext: Download fulltextPDF
Generation of Vector Images using Evolutionary Algorithms
Drázdová, Zuzana ; Pilát, Martin (advisor) ; Křen, Tomáš (referee)
The usage of evolutionary algorithms for generating images has been researched for several decades now. The potential of this approach comes from the creative power of genetic operators and broad possibilities for automated evaluation of solutions. Individuals can be either evolved to resemble an existing image or other criteria such as artistic quality can be employed. Generating vector images to resemble raster models got a lot of attention in past years. It offers several benefits. Such images can be easily scaled without any loss of accuracy. Another advantage is the option to modify individual objects in an image separately. This aspect was, so far, being neglected. We want to reach full potential of evolved images by designing a suitable algorithm. Our method generates vector images similar to given raster model that are easily editable and have an interesting artistic overlap. We developed three techniques which differ in approach to individual representation, genetic operators, evaluation and overall style of results.
Active Learning for Image Classification
Lorenzová, Kateřina ; Pilát, Martin (advisor) ; Křen, Tomáš (referee)
The thesis is a practical application of image analysis and classification methods, inspired by the behaviour of human eye. The method used doesn't require complete information of the analysed image. It uses autonomous agents that see only a part of the image instead. These agents themselves decide which other parts of the image they need to see for more precise result. The behavior of agents is controlled by a neural network, trained specifically for that purpose by an evolutionary learning algorithm. Data used for the training come from MNIST (2011), a database of handwritten numbers. This collection also contains a separate testing set, on which the behavior of autonomous agents was tested afterwards.
Deep neural networks and their application for economic data processing
Witzany, Tomáš ; Mrázová, Iveta (advisor) ; Křen, Tomáš (referee)
Title: Deep neural networks and their application for economic data processing Author: Bc. Tomáš Witzany Department: Department of Theoretical Computer Science and Mathematical Logic Supervisor: Doc. RNDr. Iveta Mrázová, CSc., Department of Theoretical Com- puter Science and Mathematical Logic Abstract: Analysis of macroeconomic time-series is key for the informed decisions of national policy makers. Economic analysis has a rich history, however when considering modeling non-linear dependencies there are many unresolved issues in this field. One of the possible tools for time-series analysis are machine learn- ing methods. Of these methods, neural networks are one of the commonly used methods to model non-linear dependencies. This work studies different types of deep neural networks and their applicability for different analysis tasks, including GDP prediction and country classification. The studied models include multi- layered neural networks, LSTM networks, convolutional networks and Kohonen maps. Historical data of the macroeconomic development across over 190 differ- ent countries over the past fifty years is presented and analysed. This data is then used to train various models using the mentioned machine learning methods. To run the experiments we used the services of the computer center MetaCentrum....
Melody generation using a genetic algorithm
Helikar, Matouš ; Maršík, Ladislav (advisor) ; Křen, Tomáš (referee)
Music composition, as all other creative activities, requires original inspiration, which can also come from melodies generated by a computer. This thesis describes generation of music tracks represented by tree structures with pluggable modules that create or alter individual musical motives. The trees can subsequently be combined by a crossbreeding algorithm driven by user ratings. This results in music tracks evolving on multiple levels, such as the selected instruments or musical motives, rhythm and overall structure. Appropriate settings of parameters for the generator and constituent modules can then produce varied tracks for inspiration or relaxation. The thesis is accompanied by a complete application using these techniques for music generation and a user study of satisfaction with the resulting tracks. Powered by TCPDF (
Maintainable type classes for Haskell
Farka, František ; Pudlák, Petr (advisor) ; Křen, Tomáš (referee)
In this thesis we address a long-term maintainability problem in Haskell type class system. In particular we study a possibility of backward-compatible changes in existing class hierarchies. In the first part of the thesis we give a brief overview of the language. The following part summarizes current proposed solutions to the problem and analyzes their properties. Based on this analysis we derive our own language extension proposal. In the penultimate chapter we present several possible applications of the language extension and compare the extension to other solutions. As a part of the thesis we also give a proof-of-concept implementation of the extension for the GHC compiler, which is briefly described in the last part of this thesis. Powered by TCPDF (
Typed Functional Genetic Programming
Křen, Tomáš ; Pudlák, Petr (advisor) ; Kubalík, Jiří (referee)
In this thesis is presented design and implementation of a system performing genetic programming in simply typed lambda calculus. Population initialization method based on term generating technique producing typed lambda terms in long normal form is introduced. This method is parameterized by simple search strategy. Several search strategies are presented, such as strategy for systematic generation or strategy corresponding to standard ramped half-and-half method. Another such a strategies called \textit{geometric} strategy is further examined in experiments and shown to have various desirable effects such as improved success rate, lesser time consumption and smaller average term size in comparison with standard ramped half-and-half generating method. Other performance enhancements are proposed and supported by experiments such as eta-normalization of generated individuals and @-tree representation of individuals. Abstraction elimination is utilized to enable use of simple tree- swapping crossover. Powered by TCPDF (
Tool for programming in a physical environment
Křen, Tomáš ; Hnětynka, Petr (advisor) ; Ježek, Pavel (referee)
The subject of this work is to implement the game conceived as an interactive physical environment in which a user creates a virtual world in hierarchical two-dimensional space by inserting, moving and connecting objects. However, the world, or his parts, also represents the syntax of a program. This is achieved mainly because the game includes different kinds of objects called functions, which occupy the same role, as a function in classic programming languages. The program also includes active agents controlled by an internal program, which is made up of these functions.

Interested in being notified about new results for this query?
Subscribe to the RSS feed.