|
|
Temperature Profiles Measurement of BGA Packages in Reflow Soldering
Tomčáková, Anna ; Bureš, Tomáš (referee) ; Starý, Jiří (advisor)
This graduation thesis addresses questions to thermal profile measurement of PBGA package during solder reflow process. The first part of thesis deals with problem of reflow process and reliability factors of solder joint connection. Next part analyses operation principles of thermocouples that are commonly used for temperature measurement. The experimental part deals with methods of thermocouples fixation during tests and measurements of dummy PBGA package. There was realized a method of dummy PBGA thermal profiles measurement and sample testing with and without simulated thermal load on PBGA package. The end of thesis concerns on possibilities of thermal profiles evaluation by using PWI method and thermal profile optimization of reflow process.
|
|
|
GUI for Active Learning of Image Detection and Classification
Bureš, Tomáš ; Šůstek, Martin (referee) ; Rozman, Jaroslav (advisor)
With active learning, domain expert doesn't need to annotate the whole dataset, but only those which will allow incremental training of a given model. An example of active learning could be detection and removal of wrong annotations. Another example is detection and expansion of training data which model fails to predict. Description of libraries, frameworks and programs which can be used to integrate with active learning is included in this work. The main part of this work is the design and description of a user interface for active learning. The application allows user to browse dataset, sort annotations and images by multiple criteria and modify annotations generated by active learning model. The application's graphical user interface was implemented with the Vue.js framework and Paper.js library. In conclusion, functionality and future application expansion are discussed.
|
|
|
Active Learning with Neural Networks
Bureš, Tomáš ; Kolář, Martin (referee) ; Hradiš, Michal (advisor)
The topic of this thesis in active learning in conjunction with neural networks. First, it deals with theory of active learning and strategies used in real life scenarios. Followed by practical part, experimenting with active learning strategie and evaluating those experiments.
|
|
|
Distributed job execution in IVIS Framework
Vašut, Roman ; Bureš, Tomáš (advisor) ; Horký, Vojtěch (referee)
This thesis tackles computation distribution in the IVIS data processing and visualization framework. In the existing versions, so-called Jobs are be- ing executed only on the IVIS host machine, raising scalability concerns. The thesis attempts to allow the distributed execution on manually-provisioned machines, commercial cloud platforms, and an HPC cluster. It does so by introducing the "executor" entity, ensuring adherence to the present Job ar- chitecture and, because the communication is done over the Internet, security. We introduce two auxiliary applications which manage the remote control of a machine and the management of a set of machines (a pool). We achieve parallelization of running Jobs. We also see the possibility of further exten- sion to enable the usage of specialized hardware or more dynamic machine allocation.
|
|
|
Data logging and visualization for Mailtrain using IVIS
Štrobl, Filip ; Bureš, Tomáš (advisor) ; Kofroň, Jan (referee)
Mailtrain is a self-hosted, free and open-source newsletter application with advanced options for managing lists of subscribers, creating and sending e-mail campaigns, and managing multiple users with granular permissions and flexible sharing. The application lacks good options for analyzing and visualizing its data for the purposes of tracking performance or security. IVIS is a framework offering the data processing and visualizing tools Mailtrain needs, and the two projects share many key technologies. In this thesis, we extend Mailtrain so that it uses IVIS and its services for logging, visualization, and analysis of its data. An emphasis is given to the extensibility of both the data logged from Mailtrain and the ways it is visualized. 1
|
|
|
Horizontal scalability for e-mail delivery in Mailtrain
Kučák, Erik ; Bureš, Tomáš (advisor) ; Kofroň, Jan (referee)
Mailtrain is a self-hosted open-source newsletter application built on Node.js which provides features such as subscriber lists management, list segmentation, custom fields, e-mail templates, triggered and RSS campaigns, etc. One of the main shortcomings of Mailtrain is the inability to scale horizontally, which results in performance limits when delivering campaigns to very large mailing lists. The main goal of this work is to extend Mailtrain to allow it to handle the delivery of campaigns (including attachments, linked images, and user tracking) in distributed and horizontally scalable manner. The thesis should include the design of the extension, its implementation, and performance evaluation to compare the extension with the existing performance of Mailtrain.
|
|
|
Machine-learning-based self-adaptation of component ensembles
Töpfer, Michal ; Bureš, Tomáš (advisor) ; Parízek, Pavel (referee)
In the area of distributed self-adaptive smart systems (such as applications of Internet of Things and Cyber-Physical Systems), machine learning has been successfully used in several applications including the prediction of metrics regarding the components in the system (e.g., battery consumption), and pruning of the space of possible adaptations. It is clear that machine learning can be a useful tool in self-adaptive systems. Most of the research works focus on using the machine learning algorithms for a specific task, yet they are (at least partially) lacking in providing a systematic approach to the introduction of machine learning into the architecture of the system. In this thesis, we propose ML-DEECo - a machine-learning-enabled component model for adaptive component architectures. It is based on the concepts of autonomous com- ponents and their ensembles (coalitions) from the DEECo component model. We enrich DEECo with abstractions for specifying machine-learning-based estimates directly in the architecture of the system. The architect can thus focus on the business logic of the application while all the tasks necessary to provide the estimates (such as collecting the data and training the model) are provided by our runtime framework. We provide an implementation of the ML-DEECo runtime in Python and...
|
|
|
Extensible Collaborative Development Platform
Halaša, Michal ; Hnětynka, Petr (advisor) ; Bureš, Tomáš (referee)
Title: Extensible Collaborative Development Platform Author: Michal Halaša Department / Institute: Katedra distribuovaných a spolehlivých systémů Supervisor of the bachelor thesis: RNDr. Petr Hnětynka Ph.D., S 212, Malostranské nám. 25, Praha Abstract: Currently, there exist a number of collaborative development platforms (also known as "forges"). These platforms offer to developers a set of tools for collaborative development (e.g. VCS tools, bug-tracking tools, mailing-list managers, etc). On the other hand, there exist many standalone tools for bug-tracking, etc. Usually, these standalone tools offer more features to the developers, however they have to be set up and managed separately. The goal of this thesis is to develop a collaborative development platform (forge) that is fully extensible, i.e. it will allow developers to easily add existing tools to it and manage them in a unified way. Keywords: java, forge, extensible, development platform
|
|
|
Optimizing performance of software connectors code generator
Petřek, Pavel ; Bureš, Tomáš (advisor) ; Bulej, Lubomír (referee)
Software connectors are intermediary entities used to model and realize communication in component systems. In addition to the communication, connectors can also provide extra functionality, such as logging or monitoring. This variability requires generation of the connector's code according to valid functional and non-functional requirements. Some requirements cannot be specified sooner than at deployment-time. However, the deployment environment can be restrictive. The existing connector generator [32] utilizes complex external tools for generation of the connector's classes from templates. In this thesis, we propose an optimization based on a precompilation approach. Templates are precompiled at design-time into a form that can be later compiled easily using a bytecode manipulation technique.
|
|
|
Auction
Mencl, Vojtěch ; Kocanda, Jiří (advisor) ; Bureš, Tomáš (referee)
Trading on the Internet is a modern way of offering products and services. Electronic auction belongs among interesting ways of trading on the Internet. The aim of this work is to analyze, design and implement e-commerce. Inaukce is a web application which convey auction or bazaar sale of various items on the Internet. Users access the application in different roles using web browser. These roles differ in users rights. After being logged on, users with appropriate rights can offer or buy items. In the administrative role, there is a option to manage user accounts and traded items. System monitoring is also available.
|
guest ::
