|
|
iOS emulator for Windows
Joneš, Jan ; Kratochvíl, Miroslav (advisor) ; Kofroň, Jan (referee)
The goal of this thesis is to create a program for Windows that takes a compiled iOS application and emulates it. However, only the application's machine code is emulated, whereas system functionality originally provided by iOS is translated to an equivalent functionality available on Windows. Hence, the emulated application employs a user interface and behavior that feel native on the target platform. At compile time, custom machine code is generated that supports the translation at runtime. The thesis also describes iOS's internals that the emulator needs to imitate and discusses different approaches to cross-platform development. 1
|
|
|
Traffic sign classification by deep learning
Harmanec, Adam ; Blažek, Jan (advisor) ; Kratochvíl, Miroslav (referee)
Classification of road signs has been studied for many years and very promising results have been achieved. We present the analysis of used data sets as very limited for real case classification. In this thesis we analyse publicly available data sets and by merging and extending them, we create a wider and more comprehensive data set applicable in the Czech Republic. Finally, we propose a new convolutional neural network architecture and test it along with several preprocessing techniques on the new data set reaching accuracy of over 99%.
|
|
|
OCR for tabular data
Tódová, Lucia ; Kratochvíl, Miroslav (advisor) ; Šefl, Vít (referee)
Table recognition is an important tool for digitalizing documents that con- tain tabular data, which often occur in areas of administration, finances and education. This thesis re-uses existing optical character recognition software to construct a new table recognition algorithm that aims to simplify the digitaliza- tion of diverse document types. The resulting algorithm achieves comparable or better results than currently available open-source software. Thesis additionally reviews common methods of OCR software implementation, and measures the influence of image preprocessing quality on the outcome of the table recognition. 1
|
|
|
Accelerating structure search in small-molecule databases
Kratochvíl, Miroslav ; Bednárek, David (advisor) ; Hoksza, David (referee)
Structure search is one of the valuable capabilities of small-molecule databases. Available chemical cartridges typically provide acceptable search performance for processing user queries, but do not scale satisfactorily with dataset size. This thesis presents Sachem, a new open-source chemical car- tridge that implements a novel method of substructure search, which em- ploys newly designed fingerprints stored in inverted indexes. The perfor- mance of the method was assessed on datasets that contain tens of mil- lions of molecules. Comparison of the performance to that of other available cartridges revealed improvements in overall search speed, scaling potential and screen-out efficiency. Additionally, the thesis presents an application of Sachem; a SPARQL service that augments existing semantic services by including results of substructure and similarity searches in small-molecule databases. The result offers new possibilities for simpler querying of the interoperable heterogeneous data sources. 1
|
|
|
Frege IDE with JetBrains MPS
Satmári, István ; Parízek, Pavel (advisor) ; Kratochvíl, Miroslav (referee)
Frege is an open-source project which brings the popular functional programming language Haskell to the Java ecosystem. JetBrains MPS is an open-source language workbench which allows users to design a new language and build an integrated development environment with a projectional (structured) editor for the created language. In this work we analyzed Frege grammar and created an IDE based on MPS that assists developers with writing code in the Frege language. Our environment includes a set of intuitive editors for editing Frege syntax, provides a simple type checking and implements code generators for the Frege language. Aim of the Frege IDE is its usability. Additionally, the thesis compares projectional editors with the more common plain-text IDEs, such as Eclipse, and evaluates whether they offer any advantage for editing purely functional programming languages.
|
|
|
Adversarial examples generation for deep neural networks
Procházka, Štěpán ; Neruda, Roman (advisor) ; Kratochvíl, Miroslav (referee)
Machine learning models exhibit vulnerability to adversarial examples i.e., artificially created inputs that become misinterpreted. The goal of this work is to explore black-box adversarial attacks on deep networks performing image classification. The role of surrogate machine learning models for adversarial attacks is studied, and a special version of a genetic algorithm for generating adversarial examples is proposed. The efficiency of attacks is validated by a multitude of experiments with the Fashion MNIST data set. The experimental results verify the usability of our approach with surprisingly good performance in several cases, such as non-targeted attack on residual networks.
|
|
|
Pattern recognition for in-game spell systems
Mikuš, Pavel ; Kratochvíl, Miroslav (advisor) ; Šefl, Vít (referee)
Magic is a popular element in current computer games. Although most games spoil the sensation of magic as of something extraordinarily subtle by allowing the player to cast spells by simply hitting key combinations, several games require the player to finish a more complicated action before casting a spell: Drawing a complicated glyph that represents the spell is one of such actions. This thesis aims to provide a repurposable library that would allow simple implementation of structured glyph-drawing-based in-game spell systems. The thesis studies several relevant approaches to pattern recognition, describes a neural-network based method for recognition of various shapes and shape combinations, develops a system for describing the parameters and results of the used algorithm in terms of predefined spell shapes and their recognized combinations, and implements this approach in a library and an accompanying simple demonstrational game. The library and its parameters are benchmarked and systematically optimized.
|
|
|
A Minimalistic Directory Service
Hrubý, Ondřej ; Mareš, Martin (advisor) ; Kratochvíl, Miroslav (referee)
We present a light, simple and secure network protocol for accessing di- rectory services called Feather-weight Directory Access Protocol (FDAP). It is inspired by Lightweight Directory Access Protocol (LDAP), but the con- cept of a directory service is built from scratch. This decision is supported by analysis of shortcomings of LDAP which has seen widespread use in the past. We provide specifications both of an FDAP service and the protocol, exam- ples of well-tested server implementation, client library and an application as a proof of concept.
|
|
|
Plánovač síťového provozu pro diferencované služby
Bali, Michal ; Kratochvíl, Miroslav (advisor) ; Matěna, Vladimír (referee)
Service differentiation, the ability of the QoS-providing mechanisms to sa- tisfy different requirements of different network traffic types, is an important part of the Internet service delivery. Usual methods of improving differen- tiated service QoS require centralized traffic scheduling, which on the other hand can not react to disturbances in transit network of typical ISPs. In this thesis we describe, implement and benchmark a traffic scheduler that is simple enough to be placed at the exact bottleneck of the network where it precisely reacts to network problems; at the same time it supports a multi-flow multi-priority stochastical traffic scheduling that guarantees a level of fairness and service differentiation. The design is built on previous research in the area - it combines the ideas of CoDel with SFQ. We implement the resulting traffic scheduler, called Multilevel Stochasti- cally Fair CoDel (MSFC), in the ns-3 network simulator. Benchmarks on a simulated ISP-like network show improvements in QoS of the differentiated services in comparison with other non-centralized classless traffic schedulers.
|
|
|
Modern approach to user interfaces for e-mail
Hruška, Marcel ; Kratochvíl, Miroslav (advisor) ; Škoda, Dominik (referee)
Webmails are indisposable interfaces for working with the e-mail on the current Internet, mostly because of the simplicity of their deployment in bro- wsers and easy integration with many provider-specific features. The most important features that are partially or fully missing in current open-source webmail implementations include directory-less mail organization by tags, navigation driven by a high-performance fulltext search, and integration of time-management capabilities. This thesis describes a new open-source alter- native to advanced commercial webmails that possesses these features. The software integrates full-text search capabilities of the ElasticSearch database with current e-mail processing infrastructure on UNIX systems to create a back-end server application, which is used by a Javascript-based browser front-end. The performance of the solution is tested on a large e-mail dataset. 1
|
guest ::
