Full-time study

4 years

doc. Ing. Jiří Hošek, Ph.D.

Chairman :
doc. Ing. Jiří Hošek, Ph.D.
Councillor internal :
prof. Ing. Jaroslav Koton, Ph.D.
prof. Ing. Zdeněk Smékal, CSc.
prof. Ing. Jiří Mišurec, CSc.
doc. Ing. Vladislav Škorpil, CSc.
doc. Ing. Radim Burget, Ph.D.
Councillor external :
prof. Ing. Ivan Baroňák, Ph.D.
doc. Ing. Miloš Orgoň, Ph.D.
doc. Ing. Otto Dostál, CSc.

The student is fostered to use the theoretical knowledge and experience gained through own research activities in an innovative manner. He/She is able to efficiently use the gathered knowledge for the design of own and prospective solutions within their further experimental development and applied research. The emphasis is put on gaining both theoretical and practical skill, ability of self-decisions, definition of research and development hypotheses to propose projects spanning from basic to applied research, ability to evaluation of the results and their dissemination as research papers and presentation in front of the research community.

The doctor study program "Electronics and Information Technologies" aims to generate top research and development specialists, who have deep knowledge of principles and techniques used in communication and data wired and wireless networks and also in related areas and also in data/signal acquisition, processing and the back representation of user data on the level of application layer. The main parts of the studies are represented by areas dealing with information theory and communication techniques. The graduate has deep knowledge in communication and information technologies, data transfer and their security. The graduate is skilled in operation systems, computer languages and database systems, their usage and also design of suitable software and user applications. The graduate is able to propose new technology solution of communication tools and information systems for advanced transfer of information.

Graduates of theprogram "Electronics and Information Technologies" apply in particular in research, development and design teams, in the field of professional activity in production or business organizations, in the academic sphere and in other institutions involved in science, research, development and innovation, in all areas of the company where communication systems and information transfer through data networks are being applied and used.
Our graduates are particularly experienced in the analysis, design, creation or management of complex systems aimed for data transfer and processing, as well as in the programming, integration, support, maintenance or sale of these systems.

The doctoral studies of a student follow the Individual Study Plan (ISP), which is defined by the supervisor and the student at the beginning of the study period. The ISP is obligatory for the student, and specifies all duties being consistent with the Study and Examination Rules of BUT, which the student must successfully fulfill by the end of the study period. The duties are distributed throughout the whole study period, scored by credits/points and checked in defined dates. The current point evaluation of all activities of the student is summarized in the “Total point rating of doctoral student” document and is part of the ISP. At the beginning of the next study year the supervisor highlights eventual changes in ISP. By October, 15 of each study year the student submits the printed and signed ISP to Science Department of the faculty to check and archive.
Within mainly the first four semesters the student passes the exams of compulsory, optional-specialized and/or optional-general courses to fulfill the score limit in Study area, and concurrently the student significantly deals with the study and analysis of the knowledge specific for the field defined by the dissertation thesis theme and also continuously deals with publishing these observations and own results. In the follow-up semesters the student focuses already more to the research and development that is linked to the dissertation thesis topic and to publishing the reached results and compilation of the dissertation thesis.
By the end of the second year of studies the student passes the Doctor State Exam, where the student proves the wide overview and deep knowledge in the field linked to the dissertation thesis topic. The student must apply for this exam by April, 30 in the second year of studies. Before the Doctor State Exam the student must successfully pass the exam from English language course.
In the third and fourth year of studies the student deals with the required research activities, publishes the reached results and compiles the dissertation thesis. As part of the study duties is also completing a study period at an abroad institution or participation on an international research project with results being published or presented in abroad or another form of direct participation of the student on an international cooperation activity, which must be proved by the date of submitting the dissertation thesis.
By the end of the winter term in the fourth year of study the full-time students submit the elaborated dissertation thesis to the supervisor, who scores this elaborate. The combined students submit the elaborated dissertation thesis by the end of winter term in the fifth year of study. The final dissertation thesis is expected to be submitted by the student by the end of the fourth or fifth year of the full-time or combined study form, respectively.
In full-time study form, during the study period the student is obliged to pass a pedagogical practice, i.e. participate in the education process. The participation of the student in the pedagogical activities is part of his/her research preparations. By the pedagogical practice the student gains experience in passing the knowledge and improves the presentation skills. The pedagogical practice load (exercises, laboratories, project supervision etc.) of the student is specified by the head of the department based on the agreement with the student’s supervisor. The duty of pedagogical practice does not apply to students-payers and combined study program students. The involvement of the student in the education process within the pedagogical practice is confirmed by the supervisor in the Information System of the university.

1. Artificial Intelligence for Advanced Biomarker Imaging in Human Attention Monitoring

   The topic focuses on researching and designing advanced biomarker imaging methods using artificial intelligence for human attention monitoring. The goal is to integrate machine learning and sensor technologies to accurately analyze physiological signals related to cognitive state and concentration.

   Tutor: Burget Radim, doc. Ing., Ph.D.

2. Enhancing Large Language Model Framework to Achieve Autonomous, Self-Organized, and Decentralized Operations

   This research aims to address the limitations of current Large Language Model (LLM) frameworks, such as the presence of central managers that can create single points of failure and issues related to agents conflicting and hallucinating. While existing LLM frameworks and agents perform well for simple use cases, they struggle with handling complex tasks end-to-end. The proposed approach focuses on developing efficient and reliable AI systems capable of managing complex tasks autonomously, in a self-organized and decentralized manner. As the demand for more sophisticated and scalable AI systems grows, the limitations of centralized LLM frameworks become more apparent. Centralized systems are prone to single points of failure and can struggle to efficiently manage the complexity and scale of modern AI tasks. Moreover, issues such as agent conflicts and hallucinations (i.e., generating incorrect or nonsensical information) further hinder the reliability and effectiveness of LLMs in complex scenarios. The research objectives include developing a decentralized LLM framework that eliminates single points of failure by distributing control and decision-making among multiple autonomous agents. Additionally, it focuses on enhancing communication protocols between agents to reduce unnecessary communication and improve efficiency. Moreover, integrating Knowledge Graphs (KG), for example, to improve the explainability of LLM responses and mitigate hallucinations, and implementing Reinforcement Learning (RL) techniques to train agents on optimal communication strategies and decision-making processes. Furthermore, the goal is to create a self-organizing system capable of dynamically incorporating new agents and adapting to changing environments and tasks. The research will employ a combination of theoretical and experimental approaches to achieve these objectives. This includes designing and implementing a decentralized architecture, developing and optimizing communication protocols, utilizing RL to train agents, integrating KGs into the LLM framework, and developing mechanisms for self-organization. The expected contributions of this research include a novel decentralized LLM framework that enhances robustness and scalability, improved communication protocols that reduce computational costs and increase efficiency, enhanced explainability and reliability of LLM responses through the integration of KGs, and a self-organizing AI system capable of dynamic adaptation and continuous learning. By addressing the limitations of current LLM frameworks and developing a decentralized, autonomous, and self-organized system, this research aims to pave the way for more robust, scalable, and reliable AI solutions. This thesis is conducted in cooperation with AT&T, where supervision and assistance will be provided by AT&T to leverage their technological expertise and infrastructure, further ensuring the success and impact of this research.

   Tutor: Hošek Jiří, doc. Ing., Ph.D.

3. Evaluation and Optimization of Directional Communications Technology in On-Demand Aerial Networks

   Recently, unmanned aerial vehicles and systems attracted attention in many contexts, such as on-demand wireless connectivity provisioning. This doctoral research topic addresses the emerging air-to-everything communication, including autonomous drone interworking over air-to-air links and robust aerial networking via air-to-ground channels. It targets to evaluate and optimize this emerging technology by contributing with efficient features to improve its performance, which notably account for specific effects and behavior of directional millimeter-wave channels. The proposed radio connectivity algorithms, system architectures, and performance evaluation frameworks are expected to become of significant value toward the development of future 5G+/6G wireless systems.

   Tutor: Hošek Jiří, doc. Ing., Ph.D.

4. Exploring Novel Techniques for Fine-Tuning Large Language Models and Enhancing Retrieval-Augmented Generation (RAG)

   Abstract: This research aims to explore and develop novel techniques for fine-tuning Large Language Models (LLMs) and enhancing Retrieval-Augmented Generation (RAG) to improve the performance, efficiency, and applicability of LLMs in various complex tasks. The proposed approach includes investigating advanced fine-tuning strategies, integrating external knowledge bases, and optimizing retrieval mechanisms to create more robust and contextually aware AI systems. Additionally, the research focuses on making fine-tuning processes more efficient and reducing computational consumption, which can lead to significant cost savings. Background and Motivation: As LLMs become increasingly prevalent in various applications, the need for efficient and effective fine-tuning techniques and advanced retrieval mechanisms becomes critical. Fine-tuning LLMs can significantly improve their performance on specific tasks, while RAG enables the models to generate more accurate and contextually relevant responses by leveraging external knowledge sources. However, current techniques often face challenges related to computational costs, scalability, and the integration of diverse knowledge sources. Reducing the computational resources required for fine-tuning can lead to lower costs and make these technologies more accessible. Research Objectives: To develop innovative fine-tuning techniques that enhance the adaptability and performance of LLMs across various domains. To optimize Retrieval-Augmented Generation (RAG) by improving retrieval mechanisms and integrating diverse external knowledge bases. To investigate the use of multi-modal data (e.g., text, images, audio) in fine-tuning and RAG processes. To create efficient and scalable methods for fine-tuning LLMs that reduce computational costs and training time, leading to significant cost savings. To evaluate the impact of different fine-tuning and RAG techniques on the performance and reliability of LLMs in real-world applications. Methodology: The research will employ a combination of theoretical and experimental approaches to achieve the outlined objectives. Key components of the methodology include: Advanced Fine-Tuning Techniques: Investigate and develop new fine-tuning strategies, such as few-shot learning, meta-learning, and transfer learning, to enhance the adaptability of LLMs. Optimized Retrieval Mechanisms: Design and implement advanced retrieval algorithms that improve the accuracy and relevance of information retrieved from external knowledge bases. Multi-Modal Integration: Explore the integration of multi-modal data in fine-tuning and RAG processes to create more contextually aware and versatile AI systems. Scalability and Efficiency: Develop methods to reduce the computational costs and training time associated with fine-tuning LLMs, such as model compression and distributed training techniques, to achieve significant cost savings. Performance Evaluation: Conduct comprehensive evaluations of different fine-tuning and RAG techniques using benchmark datasets and real-world applications to assess their impact on LLM performance and reliability. Expected Contributions: The proposed research is expected to make several significant contributions to the field of AI and LLMs: Novel fine-tuning techniques that improve the adaptability and performance of LLMs across various domains. Optimized retrieval mechanisms that enhance the accuracy and relevance of Retrieval-Augmented Generation (RAG). Integration of multi-modal data to create more contextually aware and versatile AI systems. Efficient and scalable methods for fine-tuning LLMs that reduce computational costs and training time, leading to significant cost savings. Comprehensive evaluations of fine-tuning and RAG techniques, providing insights into their impact on LLM performance and reliability. Conclusion: By exploring and developing novel techniques for fine-tuning LLMs and enhancing Retrieval-Augmented Gener

   Tutor: Burget Radim, doc. Ing., Ph.D.

5. Machine learning in photonics

   Photonic systems cover a wide range of areas from data transmission, through sensors to quantum networks. Each photonic system has its own requirements for the transmission infrastructure, but also for input and output parameters. Manual optimization of large networks based on different types of signals is almost impossible. With the help of machine learning, the optimization of both the transmitted signals and the entire infrastructure can be achieved in photonic networks. Last but not least, machine learning algorithms can be used to detect and classify non-standard network behavior to minimize security risks.

   Tutor: Münster Petr, doc. Ing., Ph.D.

6. Modern fiber optic transmission systems

   Optical transmission systems are evolving very rapidly to meet the ever-increasing demands of users. In addition to data transmissions, there are also new transmissions such as exact time, stable frequency, radio over fiber, quantum signals transmission, etc. Individual types of signals have different requirements for the transmission infrastructure. Wavelength division multiplexing is now widely used to increase the capacity of optical fibers but it is necessary to address the issue of possible interference. In order to meet the requirements of future transmission systems, it is necessary to address several technical challenges, such as new optical modulation formats with high spectral efficiency, mitigation of linear and nonlinear phenomena in optical fibers, new types of optical fibers or signal amplification with minimal noise.

   Tutor: Münster Petr, doc. Ing., Ph.D.

7. New methods using artificial intelligence tools for penetration testing

   The topic is focused on research and design of new methods using artificial intelligence that can be used during the security testing (penetration test). The research is focused on suitable methods for web applications penetration testing, network infrastructure penetration testing, but also for penetration testing of dedicated devices such as smart meters. The participation on Department’s research projects is expected.

   Tutor: Jeřábek Jan, doc. Ing., Ph.D.

8. Novel distributed and quasi-distributed fiber optic sensing systems

   The work focuses on the design, simulation and development of distributed and quasi-distributed fiber optic sensing systems. These systems use conventional single-mode telecommunication optical fibers, multimode fibers, polymer optical fibers (POF), microstructural fibers, multicore fibers, or other special fibers as a sensor. Using scattering phenomena (Raman, Brillouin, or Rayleigh scattering), or possibly changing the parameters of the transmitted optical signal (change in intensity, phase, polarization, etc.), it is possible to obtain information about temperature, vibration and other physical quantities along the optical fiber.

   Tutor: Münster Petr, doc. Ing., Ph.D.

9. Optical fiber infrastructure security

   Fiber optic networks have evolved rapidly in recent years to meet the ever-increasing demand for increasing capacity. Today, optical fibers are widely used in all types of networks due to not only transmission speed or maximum achievable distance but also security. Although fiber optic networks are considered completely secure, there are ways to capture or copy part of the data signal. Both imperfections of passive optical components and, for example, monitoring outputs of active devices can be used. With the advent of quantum computers, current encryption could be broken. It is therefore necessary to address the security of fiber-optic networks, analyze security risks and propose appropriate countermeasures.

   Tutor: Münster Petr, doc. Ing., Ph.D.