|
|
Retinopathy Detection in Retina Images using Machine Learning
Kubový, Jan ; Trunda, Otakar (advisor) ; Holeňa, Martin (referee)
In collaboration with the Institute for Clinical and Experimental Medicine (IKEM) and leveraging their historical examination data, we developed a convolutional neural net- work trained to identify diabetic retinopathy from retinal images. The primary objective of this project was to establish a machine learning model applicable within the medi- cal setting of IKEM, streamlining and potentially expediting the examination process. Additionally, we designed a user-friendly website to facilitate the straightforward utiliza- tion of the trained model by physicians possessing only basic computer skills. While the neural network demonstrates good results, it is crucial to underscore its restricted adapt- ability, attributed to the compact model size and the monotonic nature of ophthalmic data sourced from a specific type of fundus camera. The proposed solution is slated for testing in a real hospital operational environment. The neural network is not intended as a replacement for the physician, but as a tool that can assist the physician in diagnostic process. 1
|
|
|
Heuristic Learning for Domain-independent Planning
Trunda, Otakar ; Barták, Roman (advisor) ; Onaindia, Eva (referee) ; Komenda, Antonín (referee)
Automated planning deals with the problem of finding a sequence of actions leading from a given state to a desired state, e.g., solving Rubik's Cube, delivering parcels, etc. The state-of-the-art automated planning techniques exploit informed forward search guided by a heuristic, where the heuristic estimates a distance from a state to a goal state. In this thesis, we present a technique to automatically construct an efficient heuristic for a given planning domain. The proposed approach is based on training a deep neural network using a set of previously solved planning problems from the same domain. We use a novel way of extracting features for states which doesn't depend on usage of existing heuristics. The trained network can be used as a heuristic on any problem from the domain of interest without any limitation on the problem size. Our experiments show that the technique is competitive with popular domain-independent heuristic. We also introduce a theoretical framework to formally analyze behavior of learned heuristics. We state and prove several theorems that establish bounds on the worst-case performance of learned heuristics.
|
|
|
Introduction of Castes in Evolution of Artificial Beings
Trunda, Otakar ; Hric, Jan (advisor) ; Holan, Tomáš (referee)
In present work we study questions about introduction of castes in evolution of artificial beings. We present the concept that allows defining various castes and its mutual relations and simulating life of members of these castes in virtual environment. We describe features of our concept and illustrate its usability on several typical scenarios using castes. In this environment we study model situations where the beings in virtual world play different roles. We observe the simulations and its results trying to learn what impact the initial conditions, agents parameters (such as Life expectancy) and environment parameters (e.g. mutation probability) have on the outcome of the simulation.
|
|
|
Using Cellular Automata for Data Encryption
Dvořák, Martin ; Trunda, Otakar (advisor) ; Mráz, František (referee)
Cellular automata are discrete systems with very simple rules but very diverse behaviour. Some cellular automata can generate high-quality pseudorandom bit sequences. This leads us to the question of whether cellular automata could be used in cryptography, as a replacement for stream ciphers for instance. We will create and compare various methods for generating long one-time-pads from short keys, where our methods will utilize cellular automata. Besides direct design of cryptographical algorithms, we will also create an evolutionary algorithm, which will try to connect our building blocks in the best possible way. The outcome of our work will be a Windows desktop application for file encryption. Powered by TCPDF (www.tcpdf.org)
|
|
|
Monte Carlo Techniques in Planning
Trunda, Otakar ; Barták, Roman (advisor) ; Toropila, Daniel (referee)
The Monte Carlo Tree Search (MCTS) algorithm has recently proved to be able to solve difficult problems in the field of optimization as well as game-playing. It has been able to address several problems that no conventional techniques have been able to solve efficiently. In this thesis we investigate possible ways to use MCTS in the field of planning and scheduling. We analyze the problem theoretically trying to identify possible difficulties when using MCTS in this field. We propose the solutions to these problems based on a modification of the algorithm and preprocessing the planning domain. We present the techniques we have developed for these tasks and we combine them into an applicable algorithm. We specialize the method for a specific kind of planning problems - the transportation problems. We compare our planner with other planning system.
|
|
|
Finding Minimum Satisfying Assignments of Boolean Formulas
Švancara, Jiří ; Balyo, Tomáš (advisor) ; Trunda, Otakar (referee)
In this thesis we examine algorithms and techniques used for solving Boolean satisfiability (SAT). Then we inspect the possibility to use them in solving the weighted short SAT problem, which is a generalization of the satisfiability problem. Given that each variable has a weight, this generalization is the problem of finding a satisfying truth assignment while using the smallest sum of weights. To solve this problem, we introduce three truth assignments of variables - True, False and Unassign. We show that not all algorithms and techniques used in modern SAT solvers can be used in our program. Those that can be converted, will be implemented using our three truth assignments. This will yield several versions of our new solver, which will be compared. Powered by TCPDF (www.tcpdf.org)
|
|
|
Using Cellular Automata for Data Compression
Polák, Marek ; Trunda, Otakar (advisor) ; Mráz, František (referee)
In this thesis we research the possibilities of using cellular automata for lossless data compression. We describe the classification of cellular automata and their current usage. We study the properties of various types of elementary cellular automata (i.e. Wolfram rules), describe their equivalence classes, the ways of forward as well as backward simulation, we examine the rules with interesting behavior. The states provided by these rules are evaluated in terms of their orderliness (e.g. the ratio of living cells or approximation of entropy). We implement some standard compression algorithms and compare them in terms of usability for best rated states. By application of acquired knowledge we propose a new compression algorithm, test it on text and image data and compare the results with traditional compression algorithms. Powered by TCPDF (www.tcpdf.org)
|
|
|
Monte Carlo Techniques in Planning
Trunda, Otakar ; Barták, Roman (advisor)
The Monte Carlo Tree Search (MCTS) algorithm has recently proved to be able to solve difficult problems in the field of optimization as well as game-playing. It has been able to address several problems that no conventional techniques have been able to solve efficiently. In this thesis, we investigate possible ways to use MCTS in the field of planning and scheduling. We analyze the problem theoretically trying to identify possible difficulties when using MCTS in this field. We propose the solutions to these problems based on a modification of the algorithm and preprocessing the planning domain. We present the techniques we have developed for these tasks and we combine them into an applicable algorithm. We specialize the method for a specific kind of planning problems - the transportation problems. We compare our planner with other planning system.
|
|
|
Framework for development of optimization algorithms
Hurt, Tomáš ; Trunda, Otakar (advisor) ; Hric, Jan (referee)
The aim of the thesis is to design and implement an efficient tool for research and testing of algorithms of the combinatorial optimization. The domain of the planning research will be explained and the steps of design and implementation of such program will be covered. The framework will support two primary for- malisms for the description of optimization problems (PDDL, SAS+ ). The inputs processing will be provided, suitable data structures and efficient implementati- ons of search algorithms will also be included. The emphasis will be on a proper object design and easy extensibility for the future development. To achieve this goal, proven principles of software engineering will be used. 1
|
|
| |
guest ::
