Šimon Mandlík, vítěz IT SPY 2020

Not a waste of bits: The best IT Master’s thesis gives artificial intelligence access to dirty data and improves machine learning to help fight viruses

16. 12. 2020
Press release, Prague, 15 December 2020

The winners of the eleventh year of the prestigious annual IT SPY competition have been announced. The Master’s thesis that won the grand prize introduces a breakthrough in the efficiency of machine learning thanks to the ability to better process large amounts of “dirty data” and was presented by Šimon Mandlík from the Faculty of Electrical Engineering at the Czech Technical University (CTU) in Prague. The second-best thesis was done by Juraj Síč from the Faculty of Informatics at Masaryk University in Brno, who designed an application that significantly simplifies and accelerates the creation of combinational circuits. Third place went to Slovakia, to Patrik Bak from the Faculty of Science at Pavol Jozef Šafárik University (UPJŠ) in Košice for developing software that is the first in the world to be able to generate sufficiently difficult geometry problems for mathematical competitions. A jury of 19 judges, composed of leading Czech academics and businesspeople, selected the winners out of more than 1,400 Master’s theses from the Czech Republic and Slovakia.

Šimon Mandlík presents a framework with the potential to dramatically improve the efficiency of artificial intelligence in very specific industries, which he successfully tested in cooperation with Cisco and Avast. He focused on a field in which the evaluation of user data was difficult due to the number of specific data formats and their volume, and where the selection very often runs only selectively and automated processing is very limited. The project has found a way to automate data evaluation and make a significant contribution to better and faster security software updates in the future. In his thesis, the author demonstrates how to faster and better protect users from new malware and security threats.

“Large companies today own huge volumes of data, which are often among their most valuable assets. The fact that processing some data sources is very difficult and thus doesn’t occur at all represents a huge missed opportunity that I wanted to change with this program, which is unique in its functionality. The moment new malware appears, all you have to do is run a script that goes through the available data and trains a new detection model, which allows more accurate suggestions for security program updates,” says Šimon Mandlík. “We tested the software with Cisco and Avast with whom we are further developing it, so the projects presented are mainly focused on cybersecurity, but it will be used much more broadly in the future.”

“Artificial intelligence and machine learning are among the fields that are significantly changing the ordinary world more and more. The winning thesis is focused on a problem involving huge volumes of data. The author presented an interesting solution whose flexibility has been demonstrated in various experiments. The overall benefits are further enhanced by the verification of practical data from large companies,” states Professor Mária Bieliková, managing director of the Kempelen Institute of Intelligent Technologies and chair of the IT SPY competition jury.

In second place was Juraj Síč from the Faculty of Informatics at Masaryk University in Brno, who created an application that streamlines the design process of combinational circuits. These are one of the basic building blocks of electronics and are found, for example, in processors or graphics cards. This application can verify the equality of incomplete circuits. In practice, based on the parameters of the circuit, even despite missing data, it can determine whether the procedure leading to the resulting circuit is correct. When designing the circuit, developers can already verify if they are doing the right thing and detect any errors at the outset. This is one of the best solutions available in the entire world, as the author confirms:

“My application successfully passed one of the most difficult tests when I had the opportunity to compare it with similar solutions at the international competition QBF Evaluation 2020. All of the competitors received the same input data on which they performed calculations, and my application won by a long shot, which shows its quality,” says Juraj Síč.

“The focus of the thesis and the scope of its processing, as well as its verification in competition against international teams, made the Master’s project a hot candidate for excellent ranking. The field dealt with in the thesis has clear uses in practice, and it is amazing that the Czech Republic and Slovakia have left their mark on its development,” states Bohumír Zoubek, director of products and services at Profinit, one of the largest software houses in the Czech Republic and a co-organizer of the competition.

Third place went to Patrik Bak at the Faculty of Science at UPJŠ in Košice. His Master’s thesis introduces the first software in the world to be able to automatically generate sufficiently complex and interesting geometry problems that can be used in mathematical Olympiads and other competitions. The creation of sufficiently difficult and at the same time elegant geometry problems is a demanding process that requires not only expertise but also creativity and lots of time. The software has even successfully passed a test, which the author describes:

“We successfully tested the software on 4,500 processors that worked a total of 40,000 hours and generated 100,000 geometry problems. The quality of my program has been confirmed by the fact that two of my problems were among the eight best geometry problems designed for the International Mathematical Olympiad and will be used not only in the prestigious Iranian Geometry Olympiad, in which more than 50 countries take part, but also in other international, Czech and even Slovak competitions,” says Patrik Bak.

“Mathematics is very important to almost all aspects of life in that it trains us in abstract thinking. At a time when we are dealing with shortcomings in Mathematics instruction, practically already in primary schools, the thesis is, although it might not be entirely obvious at first glance, in its essence a very important solution for everyone who runs mathematical Olympiads. The availability of the algorithm can, as a result, contribute to a greater popularization of these kinds of problems and, hopefully, also interest in solving them,” states Professor Mária Bieliková.

Zuzana Tocimánková from the Faculty of Electrical Engineering and Informatics at the Technical University of Košice won this year’s People’s Choice Award with her system that based on a comparison of input data on cardiac patients can assess the severity of coronary artery occlusion and can help doctors determine whether and how quickly a patient needs to be sent for an examination. This thesis was done and experimentally tested in cooperation with the East Slovak Institute of Cardiovascular Diseases.

Not only the jury of academics and businesspeople but also Jan Šeda, shopfloor IT architect at ŠKODA AUTO, the main sponsor of the IT SPY competition, agreed that all the final theses this year were at a very high level: “I was really excited about the level and theoretical and practical sides of this year’s theses because these are really very well-crafted ideas. ŠKODA AUTO has been cooperating with universities for a long time and offers a variety of development programs and internships to students and graduates from around the world, during which we help them to, among other things, create a final thesis. All the final theses have only confirmed what we already knew from experience—that Czech and Slovak IT students are some of the best in the world.”

Bohumír Zoubek from Profinit, which co-organizes IT SPY every year, sees great added value in the close cooperation between universities and the private sector:

“The cooperation of companies with students and academic institutes on final theses plays an increasingly important role. Thanks to this cooperation, students not only gain insight into how things work in practice outside of school but, most importantly, they get the opportunity to solve very specific problems using the latest scientific knowledge. And most of the work done this way can then, of course, be further developed.

At Profinit, we are aware of students’ growing interests, and every year we offer many topics in the fields of software engineering, data processing and artificial intelligence. Each of these together with student talent, knowledge and enthusiasm can lead to exceptional work that will move our field forward.”

The IT SPY competition was held for the eleventh time this year. The participants were 1,468 graduates from the 2019/20 academic year from 16 IT faculties in the Czech Republic and Slovakia. In addition to academic recognition, the winner also received 1,000 EUR for further development of the project.

You can have a look at all the projects at https://www.itspy.cz/en/galerie-nejlepsich/.

 

Complete results:

Grand Prize

Šimon Mandlík, Faculty of Electrical Engineering, Czech Technical University in Prague

Mapping the Internet: Modelling entity interactions in complex heterogeneous networks

The winning thesis significantly contributes to the further expansion of machine learning, which is often hampered by unrealistic assumptions about the amount and accuracy of input data, which prevents the practical use of artificial intelligence. Therefore, Šimon Mandlík created a versatile data library called HMill (the Hierarchical Multi-Instance Learning Library) for his Master’s thesis, which solves the given problem. And in cooperation with Avast and Cisco, he tested it on three different tasks in the field of cybersecurity. His solution shows how to detect dangerous computer programs and suspicious domains and how to easily identify devices over the network. The main benefits of this new approach are its flexibility and versatility, which allows the modelling of different data sources with the same tool without needing to use specialised methods and without compromising the accuracy of the learned models. This technology will help both researchers and companies that work with large volumes of data.

 

Second Place

Juraj Síč, Faculty of Informatics, Masaryk University in Brno

Satisfiability of DQBF using binary decision diagrams

Combinational circuits are one of the basic building blocks of the computer circuits found in processors and graphics cards. To make their design more efficient, it is necessary to detect errors in the early stages of assembly. This can be done by solving the problem of verifying the equality of incomplete combinational circuits. Juraj Síč has developed a Master’s thesis on a precise topic, which deals with the creation of a tool (application) that uses decision diagrams to determine whether a given Boolean formula is valid or not. One of the examples of how Juraj’s application can be used is a solution for the abovementioned problem. This algorithm was implemented in a DQBDD solver, and Juraj has experimentally verified that it is better than similar solvers. This was also proven at the international competition QBF Evaluation 2020, where it defeated the other tools by a large margin.

 

Third Place

Patrik Bak, Faculty of Science, Pavol Jozef Šafárik University in Košice

Automated generation of planar geometry Olympiad problems

As part of his Master’s thesis, Patrik Bak created and implemented a method that is the first in the world to generate sufficiently difficult geometry problems for use in various competitions. Creating such software is not easy at all because the main obstacle is the billions of possible geometry problems from which only those that are sufficiently difficult and complex can be selected. The software was tested in an extensive experiment using 4,500 processors in which it generated 100,000 geometry problems. Some of them have already been used in international mathematics competitions.

 

Fourth–Tenth Place (in alphabetical order)

Roman Andriushchenko, Faculty of Information Technology, University of Technology in Brno             

Computer-aided synthesis of probabilistic models

This thesis significantly accelerates and refines probabilistic programs. These play an important role in various areas of engineering and actually model programs that have given specifications (results) but must identify missing parameters to achieve these results. These programs are used, for example, to design communication protocols that handle the transfer of a certain amount of data or to find optimal strategies such as for managing energy use.

Tomáš Anlauf, Faculty of Electrical Engineering and Computer Science, VSB – Technical University of Ostrava

Multivariate network analysis

For this thesis, a computer program was created that, in contrast to other methods, significantly simplifies the analysis of multivariate networks and examines the relationships and properties of objects in them and is also interdisciplinary in use. Other disciplines could include social media, oceanography and systems and mechanical engineering.

Matej Králik, Faculty of Mathematics, Physics and Informatics, Comenius University in Bratislava

Multi-sensor accelerometer-based gesture recognition

The author of this thesis built and perfected a prototype of gloves with wearable sensors that can be connected to software and used to control industrial machines or smart homes. When doing his thesis, he tested datasets with more than ten thousand gestures and achieved very promising results for the future use of these gloves.

Karolína Kuchyňová, Faculty of Mathematics and Physics, Charles University

User identity verification based on behavioral characteristics

This thesis deals with the recognition of users in the online environment with the help of machine learning. The system learns to recognize a specific user via mouse movements and the user’s behaviour on a specific website. Based on this data, it can evaluate whether another person has logged in to the account and then log that person out or notify the account owner, which is very useful, for example, in online banking or for email clients.

Petr Nevyhoštěný, Faculty of Information Technologies, Czech Technical University in Prague

A modular and extensible tool for software fault localization

This Master’s thesis introduces an automated tool that helps developers find bugs in software, because locating them in the code is otherwise very demanding. The big benefits are that this solution can be used regardless of the programming language, it is user-friendly and it is easily extensible for new approaches.

František Pártl, Faculty of Applied Sciences, University of West Bohemia 

Design of a hashing algorithm for biometric user signatures

For this Master’s thesis, an algorithm was created that, based on a short video of a voice recording, can safely identify a given person and is not surprised by significant similarities in siblings or the same vocal colour. It can be used for a wide variety of purposes such as replacing a signature when delivering shipments, controlling access to company premises and verifying the identity of ticket holders.

Zuzana Tocimáková, Faculty of Electrical Engineering and Informatics, Technical University of Košice        

Case-based reasoning in the diagnosis of cardiovascular diseases

This thesis was done in cooperation with the East Slovak Institute of Cardiovascular Diseases and proposes a solution that allows doctors to find the known cases that most closely correspond to the condition of the current patient. Thus, it can help cardiologists assess the severity of a coronary artery occlusion and determine whether it is necessary to send the patient for a coronarographic examination. This system uses machine learning methods and has been experimentally tested.

 

 

People’s Choice

A prize awarded on the basis of online votes from the public

Zuzana Tocimáková, Faculty of Electrical Engineering and Informatics, Technical University of Košice        

Case-based reasoning in the diagnosis of cardiovascular diseases