Computer and information sciences

1. Application of quantum computers to computationally expensive problems.

Supervisor: dr hab Maciej Malawski

Auxiliary supervisor: dr inż Katarzyna Rycerz

Faculty of Computer Science, Electronics and Telecommunications

Abstract: Currently, quantum computing is one of the most dynamically developing field in modern computer science . The proposed PhD topic includes, in particular, assessment of possible applications of existing and publically available quantum computer prototypes like IBM-Q and D-Wave to computationally intensive problems such as:

- task scheduling on computing infrastructures (including optimization by D-Wave quantum annealing or optimization methods available in Quantum Information Science Kit from IBM

- application of quantum random walks for image segmentation

- research on possible implementations of quantum games on IBM-Q computer or using quantum network simulators such as SimulaQron.

Research facilities: This topic will be realized in cooperation with the Institute of Theoretical and Applied Informatics of the Polish Academy of Sciences in Gliwice, the University of Notre Dame and IBM Software Lab Kraków. Its partial results will be discussed during the Krakow Quantum Informatics Seminar organized by the Department of Computer Science and IBM lab in Krakow. Access to IBM-Q and D-Wave is possible free of charge.

Number of places: 1

 

2. Interpretability problems in deep learning models.

Supervisor: prof. dr hab. inż. Witold Dzwinel

Auxiliary supervisor: dr inż. Marcin Kurdziel

Faculty of Computer Science, Electronics and Telecommunications

Abstract: In the era of deep learning when the resulting data models are amazingly accurate, they become more and more complex to be interpretable by human. Without a methodology, which would be able to infer how a black box takes its decision, the highly effective machine learning methods in the socially and technologically sensitive domains will remain too risky to be implemented. Consequently, the lack of interpretability may be the principal issue in deploying these models to the real-world problems, particularly, the mission-critical applications such as healthcare, where being able to validate, trust and understand a learned model is an indispensable factor of its clinical implementation. Therefore, the trend for exploitation of the power of deep learning technology via their scaling towards larger and more complex systems, which will be more versatile, independent and accurate, is confronted with the loss of their interpretability. This can dramatically shrink the possibility of reasonable control and prediction of their behavior. Our project focuses on these two seemingly antagonistic aspects. However, the problem should be formulated in a much wider context. Namely, how the interpretability of the NN system depends on its complexity? To this end First, we explore the scaling opportunities of neural networks models by using ensembling strategy. Second, we try to develop the tools, allowing for both visual and algorithmic interpretation of the behavior of these NN ensembles.

Research facilities:

Number of places: 1

 

3. Authorship identification and plagiarism detection with neural networks and deep learning.

Supervisor: prof. dr hab. inż. Jacek Kitowski

Auxiliary supervisor: dr inż. Marcin Kuta

Faculty of Computer Science, Electronics and Telecommunications

Abstract: Examination of authorship finds application in digital forensics (identification of suicide attempters, terrorists, etc.), authorship attribution of literary works or authorship identification of malware. A related problem is plagiarism detection. Recent development of deep learning and neural architectures opens a new perspective for solving these problems. Theme of research includes search for new architectures of neural networks, new representations of texts based on deep networks and new algortithms for authorship identification and plagiarism detection, in order to achieve higher classification accuracy.

Research facilities: Use of the IT facilities of the Department of Computer Science and the possibility of using computational resources of the Academic Computer Center CYFRONET AGH, including the largest Polish supercomputer Prometheus.

Number of places: 1

 

4. Computing for data analysis and reconstruction for TOTEM and CMS experiments, using parallel and distributed computing.

Supervisor: dr hab Maciej Malawski

Auxiliary supervisor: dr Leszek Grzanka

Faculty of Computer Science, Electronics and Telecommunications

Abstract: Participation of Department of Computer Science in TOTEM and CMS experiments involves development of algorithms, methods and software for data analysis. In particular, the important algorithms algorithms for reconstruction of data from the detectors take into account the geometry and detection methods. Additionally, the large data volume requires the use of parallel and distributed processing methods. The proposed research area for the graduate school involves development of aforementioned tools, such as:

Development of algorithms for track reconstruction taking into account the geometry of detectors and optics of the accelerator,

Investigation of new methods and tools for big data analysis for experimental data from TOTEM and CMS,

Usage of distributed computing infrastructures such as clusters and clouds for running and optimization of processing.

Additional tasks for the PhD candidate will include: collaboration with CERN team (including shorter and longer trips), delivering results in the form of software integrated with CMSSW repository.

Research facilities: This topic will be realized in cooperation with CERN, the additional stipend for the PhD student is planned from the grant awarded by the Polish Ministry of Science and Higher Education.

Number of places: 1

 

5. AI methods for transparent data access to distributed data in HPC systems.

Supervisor: dr hab. inż. Renata Słota

Auxiliary supervisor: dr Łukasz Dutka

Faculty of Computer Science, Electronics and Telecommunications

Abstract: Scientific problems carried out jointly by international research groups require tools ensuring not only efficient access to distributed unstructured data but also easy management. Operations on metadata, representing data, environment and the user, allow to minimize data transfer to transfer only a portion of data currently modified by the user. This is done at the expense of complex operations on metadata, allowing for partial prediction of the required fragment. The use of artificial intelligence (AI) methods supporting the prediction process will allow for more efficient organization of calculations and less overhead associated with data transfer. Another issue where the use of AI methods is required is random remote access to small fragments of files that require integration before transfer to minimize the communication overhead. Due to the extensive cooperation between distributed groups of users and growing importance of Big Data issues development of effective data access methods is currently the focus of research centres.

Research facilities: Use of the IT facilities of the Department of Computer Science and the possibility of using computational resources of the Academic Computer Center CYFRONET AGH, including the largest Polish supercomputer Prometheus. Possibility to participate in international projects (e.g. under H2020) and national projects.

Number of places: 1

 

6. Scene understanding and machine learning .

Supervisor: dr hab. inż. Bogdan Kwolek, prof. AGH

Faculty of Computer Science, Electronics and Telecommunications

Abstract: The research concerns computer vision, machine learning, adaptive / real-time systems, stage perception and human-machine interaction. The scientific goal is to develop original solutions, and in particular algorithmic solutions for perception and real-time understanding of the scene. The subject of the research are computer vision algorithms, object detection, objects recognition and 6D pose estimation on digital images (RGB and RGB-D) and motion tracking. A significant part of the research concerns training of machines and adaptive systems, including using advanced machine learning techniques, designing complex models based on neural networks.

Research facilities: Published a number of works on computer vision, machine learning, real-time adaptive / learning systems, stage perception and human-machine interaction. Coordinator of seven NCN research projects. In 2018-2021 is implemented an NCN research grant "Effective methods of real-time scene perception based on multimodal data analysis and unsupervised learning of features", in which there are PhD and postdoctoral positions. Held several academic fellowships/internships: Technische Universität Münich / Bielefeld University - 26 months, Stanford University - 2 months, INRIA Paris - Rocquencourt - 1 month, University of Stavanger / University of Oslo - 6 weeks and many other short-term research visits. The team has at its disposal specialized digital cameras, a humanoid robot, GPU computing computers and specialized software.

Number of places: 2

 

7. Analysis of cancer changes in histopathological images with methods of artificial intelligence..

Supervisor: prof. dr hab. inż. Bogusław Cyganek

Faculty of Computer Science, Electronics and Telecommunications

Abstract: The goal is to conduct research into development of methods for detection and recognition of cancer changes in histopathological images of selected tissues. Application of methods of artificial intelligence, particularly deep neural networks, computer vision and machine learning is envisioned.

Research facilities: The department is in possession of necessary stuff as well as technical facilities such as computational servers, cameras, graphic cards, literature, etc. required for realization of the research task.

Number of places: 1

 

8. Algorithmic Analysis of Elections

Supervisor: dr hab. inż. Piotr Faliszewski

Faculty of Computer Science, Electronics and Telecommunications

Abstract: In this research program, the PhD student will analyze computational complexity of various problems releted to elections and voting, including the problems of winner determination, the problems of winner prediction, or the problems of manipulating election outcomes. Research work will involve creating algorithms and analyzing the computational complexity of appropriate problems, related to both single-winner and multiwinner elections. The research may involve attempts to solving real-life problems from other disciplines via election methods.

Research facilities:There is a research group in the Department of Computer Science that works on algorithmic aspects of elections. Carrying out the research in the proposed topic requires theoretical tools (such as the ability of formulating and proving mathematical theorems) and access to computational power. The necessary research facilities are provided by the department

Number of places: 1