|
|
Using genetic programming in robot evolution
Babor, Petr ; Mráz, František (advisor) ; Neruda, Roman (referee)
Artificial neural networks learned by evolutionary algorithms are commonly used to control the robots. Neural networks can be encoded either directly as a list of weights or indirectly as a weight generator. Unlike direct coding indirect encoding allows to encode a large network using a short genetic code. HyperNEAT is a neuroevolutionary algorithm, which encodes the neural network indirectly, through another (producing) network, which computes synaptic weights. A different algorithm called HyperGP is an alternative to HyperNEAT. In HyperGP, the producing network is replaced by an arithmetic expression, which is being evolved using a genetic programming (GP). We have designed enhancements for HyperGP, using techniques that are either known in a different context of GP or completely new. Algorithm and enhancements have been implemented and experimentally tested on a task of controlling virtual walking robot. The results were compared with HyperNEAT and with the original HyperGP. We have shown that most of the proposed enhancements are effective and, on the given task, HyperGP is better than HyperNEAT. GP thus can successfully replace NEAT in hyper-encoding scheme and improve its efficiency. Powered by TCPDF (www.tcpdf.org)
|
|
| |
|
|
Computational Intelligence Methods in Metalearning
Šmíd, Jakub ; Neruda, Roman (advisor) ; Vanschoren, Joaquin (referee) ; Vomlelová, Marta (referee)
This thesis focuses on the algorithm selection problem, in which the goal is to recommend machine learning algorithms to a new dataset. The idea behind solving this issue is that algorithm performs similarly on similar datasets. The usual approach is to base the similarity measure on the fixed vector of metafeatures extracted out of each dataset. However, as the number of attributes among datasets varies, we may be loosing important information. Herein, we propose a family of algorithms able to handle even the non-propositional representations of datasets. Our methods use the idea of attribute assignment that builds the distance measure between datasets as a sum of distance given by the optimal assignment and an attribute distance measure. Furthermore, we prove that under certain conditions, we can guarantee the resulting dataset distance to be a metric. We carry out a series of metalearning experiments on the data extracted from the OpenML repository. We build up attribute distance using Genetic Algorithms, Genetic Programming and several regularization techniques such as multi-objectivization, coevolution, and bootstrapping. The experiment indicates that the resulting dataset distance can be successfully applied on the algorithm selection problem. Although we use the proposed distance measures exclusively...
|
|
|
Koordinace chování virtuálních lidí
Gemrot, Jakub ; Brom, Cyril (advisor) ; Neruda, Roman (referee)
This thesis is about specific approach to the behavior coordination of multiple embodied virtual agents. Agents may act for themselves or be controlled directly by bodiless coordination agents. This kind of approach is designed for the area of interactive storytelling, where the actor agents are viewed as a string puppets that are controlled by the abstract director. The control mechanism is based upon the BDI architecture and the AgentSpeak(L) language that is extended with template plans and new plan execution mechanism that allows the directing of other actor agents.
|
|
|
The Evolution of Football Strategies
Jiřička, Martin ; Holan, Tomáš (advisor) ; Neruda, Roman (referee)
The thesis describes a design of continuous simulation of simplified football match and it also describes an evolution, that offers to a user a chance to evolve a football team, that competes best in the simulation. Both simulation and evolution have been implemented as a program for Microsoft .NET Fra- mework. Program allows simple distributed computation of an evolution and also viewing a results and watching matches in 2D graphics. So that part of the thesis is dedicated to a description of how the program works and how to use it. In the end are mentioned some performed computations illustrating how the pro- gram could be used and a small analysis that shows on concrete evolution that the designed solution mights work. 1
|
|
|
Adaptive Matchmaking Algorithms for Computational Multi-Agent Systems
Kazík, Ondřej ; Neruda, Roman (advisor) ; Paprzycki, Marcin (referee) ; Diamantini, Claudia (referee)
The multi-agent systems (MAS) has proven their suitability for implementation of complex software systems. In this work, we have analyzed and designed the data mining MAS by means of role-based organizational model. The organiza- tional model and the model of data mining methods have been formalized in the description logic. By matchmaking which is the main subject of our research, we understand the recommendation of computational agents, i.e. agents encap- sulating some computational method, according their capabilities and previous performances. The matchmaking thus consist of two parts: querying the ontol- ogy model and the meta-learning. Three meta-learning scenarios were tested: optimization in the parameter space, multi-objective optimization of data min- ing processes and method recommendation. A set of experiments in these areas have been performed. 1
|
|
|
Hyperparameter optimization in AutoML systems
Pešková, Klára ; Neruda, Roman (advisor) ; Awad, Mariette (referee) ; Kordik, Pavel (referee)
In the last few years, as processing the data became a part of everyday life in different areas of human activity, the automated machine learning systems that are designed to help with the process of data mining, are on the rise. Various metalearning techniques, including recommendation of the right method to use, or the sequence of steps to take, and to find its optimum hyperparameters configuration, are integrated into these systems to help the researchers with the machine learning tasks. In this thesis, we proposed metalearning algorithms and techniques for hyperparameters optimization, narrowing the intervals of hyperparameters, and recommendations of a machine learning method for a never before seen dataset. We designed two AutoML machine learning systems, where these metalearning techniques are implemented. The extensive set of experiments was proposed to evaluate these algorithms, and the results are presented.
|
|
|
Artificial Composition of Multi-Instrumental Polyphonic Music
Samuel, David ; Pilát, Martin (advisor) ; Neruda, Roman (referee)
David Samuel We propose a generative model for artificial composition of both classical and popular music with the goal of producing music as well as humans do. The problem is that music is based on a highly sophisticated hierarchical structure and it is hard to measure its quality automatically. Contrary to other's work, we try to generate a symbolic representation of music with multiple different instruments playing simultaneously to cover a broader musical space. We train three modules based on LSTM networks to generate the music; a lot of effort is put into reducing high complexity of multi-instrumental music representation by a thorough musical analysis. Our work serves mainly as a proof-of-concept for music composition. We believe that the proposed preprocessing techniques and symbolic representation constitute a useful resource for future research in this field. 1
|
|
|
A virtual company simulation by means of autonomous agents
Bošanský, Branislav ; Brom, Cyril (advisor) ; Neruda, Roman (referee)
Business process modeling is one of the most used approaches for capturing the work practise. However, if we want to simulate the execution of these processes, the specification based on process modeling is not suitable, because it is usually based on pure statistical calculation. On the other side there are agent-based simulations that provides more realistic model of a simulated system, but the specification of the simulation by means of agents is usually much more complicated. Thus, in this thesis, we address to the interconnection of these two different approaches and examine the possibilities of agents' behaviour definition using an enhanced process modeling language. We follow the EPC language and its epresentation in EPML, which we extend to carry informations necessary for agents, creating that way a new language A-EPML. To prove that this language is capable of being transformed onto the rules determining agents' behaviour, we present the prototype implementation that performs this translation to the final multi-agent system representing a virtual company.
|
|
|
Process Mediation Framework for Semantic Web Services
Vaculín, Roman ; Neruda, Roman (advisor) ; Nečaský, Martin (referee) ; Svátek, Vojtěch (referee)
The goal of Web services is to enable interoperability of heterogeneous software systems. Semantic Web services enhance syntactic specifications of traditional Web services with machine processable semantic annotations to facilitate interoperability. AsWeb services get popular in both corporate and open environments, the ability to deal with uncompatibilities of service requesters and providers becomes a critical factor for achieving interoperability. Process mediation solves the problem of interoperability by identifying and resolving all incompatibilities and by mediating between service requesters and providers. In this thesis we address the problem of process mediation of Semantic Web services. We introduce an Abstract Process Mediation Framework that identifies the key functional areas to be addressed by process mediation components. Specifically, we focus on process mediation algorithms, discovery of external services, monitoring, and fault handling and recovery. We present algorithms for solving the process mediation problem in two scenarios: (a) when the mediation process has complete visibility of the process model of the service provider and the service requester (complete visibility scenario), and (b) when the mediation process has visibility only of the process model of the service provider but...
|
guest ::
