ez itt az index

Stochastic systems and data modeling

Activity of the research group:

Modelling computer and communication systems with stochastic processes, in particular Markov chains, fluid models, reward models and mean-field models. Efficient numerical solution of large-scale Markov chains with regular structure. Theory of large deviations. Numerical inverse Laplace transform. Data modelling by fitting phase type processes with Markov arrival processes. Main application areas (with industrial experience): performance analysis of telecommunication systems, efficient routing in sensor networks, modelling of financial and IoT data sets, anomaly detection.

Recent results:

• We have developed network dimensioning and performance analysis algorithms based on stochastic modelling for radio access networks in 2G, 3G and 4G networks, which are being used in Nokia’s products
• We developed a new algorithm for detecting anomalies in telecommunication data sets for Nokia – Bell Labs, which won the Pro Progressio Foundation Innovation Grand Prize, the IT Business Innovation Award and the product that uses the algorithm won the 5G World Summit “Most Innovative Artificial Intelligence or Machine Learning Technology for the Network” award
• We developed fast and numerically stable algorithms for the analysis of multi-class non-Markovian queuing systems
• We provided analytical results for V2V message propagation speed in a highway environment
• Development of efficient MU-MIMO communication systems
• Laplace transforms are used in many engineering fields, but inverse transforms are notoriously difficult. Our published method based on concentrated matrix exponential functions is the best algorithm currently available for this purpose

Recent projects:

OKTA • MILAB • MTA TKI (Information Systems Research Group)

International relations:

TU Dortmund • Universita di Torino • University of Antwerp

Industrial partners:

Nokia – Bell Labs • Ericsson

Artificial Intelligence

Activity of the research group:

Our research covers the full spectrum of artificial intelligence (MI), organically linked to our group’s decades-long faculty role in general AI education, which is now present in BProf, BSc, MSc and PhD programs. Our working groups within the research group are Digital Humanities, Image Processing and Multimodal Diagnostics, Cooperative Intelligence, and Computational Biomedicine (ComBine). Digital Humanities examines semantic technologies through the processing of natural language. Image Processing and Multimodal Diagnostics explores the integration of classical image processing and information fusion methods with new artificial intelligence-based methods. Cooperative intelligence research examines federated learning, sensor networks, autonomous vehicles, multi-agent systems, and distributed computing methods. The ComBine team is developing artificial intelligence methods for bioinformatics, chemoinformatics and health informatics, with a particular focus on data and knowledge fusion, systems-based data analysis, and causal discovery.

Recent results:

We have launched an Artificial General Intelligence course to teach about the grand challenges, limits, trends, interdisciplinary nature and ethical, social dimensions of artificial intelligence. Our group is also participating in the development of a new Human-Centered Artificial Intelligence MSc program, which is part of an international EU project. We have launched and organized a student competition at the faculty in Artificial Intelligence and Machine Learning for the third time. In our research, we routinely use and constantly expand our Bayesian systems-based analysis to support biomarker discovery in several projects (UKB1602, TRAJECTOME, OTKA119866, OTKA139330). In addition to our statistical genetic research, we also focus on the development of AI methods for chemoninformatics and drug discovery, which is strengthened by our participation in national and international projects (MELLODDY). Our intelligent image processing and multimodal data fusion research has also been boosted by new projects, and we are examining the use of modern artificial intelligence tools in natural language processing in several collaborations.

Recent projects:

H2020/IMI2, MELLODDY • EUREKA, iCare4NextGen • OTKA K139330

International relations:

University of Minnesota • K.U.Leuven

Industrial partners:

SOTE • Richter • Dealogic • Abylon Kft. • E-Group • Continental

Control engineering and robotics

Activity of the research group:

Investigation, development and implementation of modelling, signal processing and control techniques used in robotics. Autonomous behaviour of robots and robot groups, navigation and trajectory planning algorithms, specification and implementation of intelligent functions based on these algorithms. Robot applications in real and virtual environments, robot control hardware and software architectures.

Recent results:

TODO

Special infrastructure:

Mitsubishi robotic arm • KUKA robotic arm

Recent projects:

VKE • H2020 • TKP/FIKP

International relations:

University of New South Wales • Kyungpook National University • University of New Hampshire • Griffith University

Industrial partners:

KUKA • Gamma digital • Knorr Bremse • ThyssenKrupp • Continental • Akytec • Emerson

Intelligent transportation systems

Activity of the research group:

The mission of the Laboratory of Multimedia Networks and Services (MEDIANETS) is to combine traditional strengths in networking with new competences in articifial intelligence, applying it to smart city environment. Research and development activities include machine learning and data analytics for intelligent and automated cities, V2X communication and intelligent transportation systems. MEDIANETS is active in significant European and national research projects, together with R&D projects carried out with our partners from the ICT industry.

Recent results:

• Optimization of warehouse logistic processes with machine learning (Cooperating partners: UPS Netherlands, Bright Cape Holdings, University of Trento, Fundazione Bruno Kessler, RISE)
• Development of an intelligent transportation system for smart cities (Cooperating partners: Gamax Kft, RacioNet Zrt)
• ETSI ITS-G5 based V2X R-ITS-S infrastructure planning for sections of the M76 and M9 highways (Cooperating partners: Utiber, Roden, NIF)
• Providing C-ROADS platform consultation services for Magyar Közút – Hungarian Public Roads
• Development of a 5G Telco Cloud Simulator (Cooperating partner: Nokia Bell Labs)
• Development of an intelligent parking system, based on machine learning (Cooperating partner: T-Systems)
• Digital profiling (Cooperating partner: Nokia Bell Labs)
• Development of a Cloud-in-the-loop simulator for automotive edge computing (Cooperating partner: Ericsson)

Special infrastructure:

Commsignia RS2/OB2

Recent projects:

VKE

International relations:

University of Babylon • Victoria University of Wellington, New Zealand • University of Cauca, Colombia • University of Donja Gorica • Technical University of Kosice, Slovakia • Raytheon BBN Technologies, University of Iowa • Universite libre de Bruxelles

Industrial partners:

Commsignia • T-Systems • RacioNet Zrt • Gamax Kft • Nokia Bell Labs • Ericsson • Utiber

Discrete mathematics

Activity of the research group:

Fundamental research in graph theory, hypergraphs, combinatorics, combinatorial optimization, combinatorial number theory, game theory, database theory, rigidity of graphs and structures, additive combinatorics, combinatorial geometry, search theory, extremal set systems, relation between databases and code theory, graph coloring, behaviour of graph parameters in product graphs

Recent results:

Péter Pach Pál developed a new version of the polynomial method in 2016 together with Croot and Lev. This new method has led to the solution of famous problems such as the cap set problem or the Erdős-Szemerédi sunflower conjeture. Since then, the method has had many applications, such as exact bound for Green’s lemma of “arithmetical triangle removal” (Fox-Lovász), Sárközy’s theorem for polynomials over finite bodies (Green), and many others. The article was published in the most prestigious mathematical journal, Annals of Mathematics, and Fields Medal-winning mathematicians Gowers, Tao, and other leading mathematicians such as Cameron and Kalai have also analyzed it on their blogs.

Géza Tóth, together with János Pach and Gábor Tardos, proved far-reaching generalizations of the Crossing Lemma to multigraphs under various natural conditions.

Gábor Wiener, together with Peter Dameschke and Azam Sheikh Muhammad, laid the combinatorial foundations of a new, practical and well-used strict group testing model, which was published in the Journal of Combinatorial Theory A, one of the leading combinatorial journals.

Gábor Simonyi, together with Gábor Tardos, gave a partial (complete in the 4-chromatic case) characterisation of the colour-critical edges of Schrijver graphs.

Gyula Katona and László Papp, in a joint work with Ervin Győri, gave lower and upper bounds on the optimal pebbling number of large grids.

Gyula Katona, with Kitti Varga, achieved several significant results in the study of minimally tough graphs.

Recent projects:

MTA Lendület • OTKA

International relations:

Ibaraki University, Japan • Lancaster University, UK • University of Haifa, Israel • University of Warwick, UK • Ghent University • Yokohama National University • University of British Columbia, Canada • Sapienza – Universitá di Roma

Industrial partners:

Morgan Stanley • Lynx Analytics

CrySyS Lab

Activity of the research group:

Our research group works in 3 domains within the field of security and privacy:

1. security of cyber-physical systems,
2. security and privacy problems in machine learning-based systems,
3. economics of security and privacy.

In domain (1), we work on the security of industrial automation and control systems, security of modern vehicles and intelligent transport systems, and security of IoT systems and applications. The common in these topics is that attacks originating from cyberspace may have physical consequences, resulting in equipment or environmental damage, or potentially even loss of human life, and therefore, security is an important requirement. In domain (2), we study how machine learning can be used to solve security and privacy problems, and also how machine learning–based systems may be exploited maliciously. More specifically, we focus on the security of federated learning algorithms and the problem of adversarial examples (e.g., in machine learning-based malware detection). In domain (3), we apply game theoretic models to study the incentive structures in different systems, and the cause of security and privacy problems. Besides the domains mentioned above, we have strong competency in applied cryptography, privacy enhancing technologies, malware analysis, reverse engineering, and secure operation of networks and network-based systems, including IT infrastructure automation.

Recent results:

• In the SETIT project, we developed an efficient anti-virus solution for resource constrained IoT devices, which detects malware with more than 90% accuracy.
• We developed many further security enhancing solutions for IoT devices.
• We constructed new attacks on the vehicle CAN bus, and proposed multiple attack detection mechanisms.
• We analyzed what information can be extracted from recorded CAN traffic, and showed that it is possible to identify the driver solely from the raw CAN data.
• We designed and implemented honeypots for ICS/SCADA systems.
• We achieved pioneering results in the field of interdependent privacy, which gave the theoretical explanations for the Cambridge Analytica scandal.
• We proposed new privacy-preserving federated machine learning algorithms.

Special infrastructure:

IoT devices, PLCs, industrial devices, servers • PIRAMID ICS/SCADA security testbed

Recent projects:

PrOTectME (EIT Digital) • H2020 MELLODDY • H2020 SECREDAS • H2020 SETIT

International relations:

NTNU, Trondheim, Norway • KU Leuven, Belgium • INRIA Rhones-Alpes, France • University of California, Irvine, CA • New York Institute of Technology

Industrial partners:

Microsec Zrt. • Tresorit Kft. • Ukatemi Technologies Kft.

Speech Communication and Smart Interactions

Activity of the research group:

Our lab group is an “old hand”, we have been present at the Department of Telecommunications and Media Informatics of the Budapest University of Technology and Economics since the beginning. Our main areas of expertise are speech communication and speech technology in Hungarian, human-machine, human-robot interfaces, assistive interfaces for blind, visually impaired and speech impaired users, general statistics and modern machine learning algorithms. A strong focus on deep learning research and education. We have gained experience in a number of national and international projects in the broader field of ICT applications (customer service automation, self-driving cars, etc.). Our technologies have been applied in practice in these areas for decades.

Recent results:

• Research:
  • H2020 projects (AI4EU partner, APH-ALARM AAL partner, VUK and PAELIFE AAL projects), MI coalition activity, member of national laboratories (MI Lab, InfoLab),
• Industry:
  • Continental Professional Intelligence for Automotive project, Idomsoft ücustomer service automation, IT.DOT MÁV voice passenger information, OTP speaking ATMs (in daily use, also in other banks besides OTP + government systems under implementation/planning)
• Education:
  • Educating academia and industry on artificial intelligence and deep learning.
  • Introduction of practical deep learning and HCI teaching at the faculty, Teaching of Speech Information Systems, Brain Signal Processing, Speech and Hearing Diagnostics, Participation in the development and teaching of Infocommunication and other subjects
  • NVidia Deep Learning Institute organisation and delivery of full-day workshops on deep learning for academics and students

Special infrastructure:

GPU servers

Recent projects:

OTKA • CELSA

International relations:

University of Novi Sad, Serbia • University of Ss. Cyril and Methodius, Skopje, Macedonia • diap Research Institute, Martigny, Switzerland • Université Pierre et Marie Curie, Paris, France • Langevin Institute, ESPCI-ParisTech, Paris, France • Tianjin University, Tianjin, China • University of Cologne, Cologne, Germany • Universidad Industrial de Santander Bucaramanga, Santander, Colombia • UPC Barcelona, Barcelona, Spain • National University of Defense Technology, Changsha, China • AGH University of Science and Technology, Department of Measurement and Electronics • University of Ss. Cyril and Methodius in Skopje, Macedonia • Università della Campania Luigi Vanvitelli and IIASS • Université Grenoble Alpes • Universidad Industrial de Santander Bucaramanga, Santander, Colombia

Industrial partners:

ELTE • SOTE • ELKH • Hospitals • Idomsoft • IT.DOT Kft • OTP • Microsoft Mo. Kft. • Auxiliis Pharma Kft. • SCI Hálózat zRt.

Applied IT technology

Activity of the research group:

The Applied Computer Science group’s research interests include software development methods, efficient and maintainable development techniques, and testing. Other research areas include scalable software architectures and environments, and applied artificial intelligence, machine learning and big data, and business intelligence solutions. This is complemented by the area of smart data collection, which is of particular importance for the design and implementation of large-scale systems.

Recent results:

TODO

Recent projects:

MI Nemzeti Labor • Autonóm Nemzeti Labor • TKP

International relations:

University of Dresden • University of Helsinki

Industrial partners:

Sagemcom • Nokia

Data Science and Content Technologies

Activity of the research group:

The group deals with modern methods of data mining, information retrieval, natural language processing, machine learning, image and video processing. These belong to the fields of artificial intelligence and data science, within which the research focuses on computer vision, neural networks, deep learning, active learning. In addition to research, the group also plays a significant role in education, as responsible for Specialization of Data Science and Media Informatics in the MSc Computer Engineering, Specialization of Analytical Business Intelligence in the MSc Business Informatics, and Data-Based Systems Specialization in BProf Computer Engineering.

Recent results:

• Active learning in image classification, Few-shot learning in X-ray image analysis,
• Character recognition (digit recognition) in extreme conditions,
• Spectral clustering on multimedia datasets, Recognition of plants in images by deep learning, Balanced active learning, Machine learning in an open set.
• 1st place in the SnakeCLEF 2021 International Image Classification Competition organized by ImageCLEF / LifeCLEF

Special infrastructure:

GPU (Titan X)

Industrial partners:

Posta • Tigáz • Magyar közút • Pannon University • ELTE • Debrecen University • NKE