Autonomous robots, neural networks, and preventing attacks on information systems: ETU "LETI" at the MECO conference
ETU "LETI" co-organized and presented 34 developments at the 9th Mediterranean Conference on Embedded Computing MECO, which took place from June 8 to 11 in Budva, Montenegro.
The organizers of the 9th Mediterranean Conference on Embedded Computing MECO were the University of Montenegro, Eindhoven University of Technology (Netherlands), European Academy, University of Zagreb (Croatia), Saint Petersburg Electrotechnical University "LETI," and Ryazan State Radio Engineering University.
Three ETU "LETI" researchers joined the scientific committee of MECO: Dmitry Kaplun, Associate Professor of the Department of Automation and Control Processes, Evgeniya Novikova, Associate Professor of the Department of Information Systems, and Ivan Kholod, Dean of the Faculty of Computer Science and Technology.
In 2020, MECO attracted a record number of participants - over 650 authors from more than 45 countries on five continents - who participated in 15 sections and 30 online sessions. These were representatives of the strongest engineering universities and leading high-tech companies: NVIDIA, Technical University of Munich, Karlsruhe Institute of Technology, Mercedes-Benz, BMW, Volkswagen, Cisco Systems, Ericsson Research, Technical University of Dresden, Eindhoven University of Technology, Vienna University of Technology, Istanbul Technical University, University of Manchester, Norwegian University of Science and Technology, Hong Kong University of Science and Technology, Saint Petersburg Electrotechnical University "LETI," King Abdullah University of Science and Technology, Technical University of Sofia, University of Antwerp, Slovak University of Technology in Bratislava, Czech Technical University in Prague, University of Patras, University of California, Irvine, Moscow Power Engineering Institute, National and Kapodistrian University of Athens, University of Siena and many others.
Kirill Krinkin, Head of the Department of Software Engineering and Computer Applications, Head of Artificial Intelligence R&D area at ETU "LETI," was a keynote speaker at the conference and presented the report "Transferable Belief Models for Lightweight Simultaneous Localization and Mapping," focused on the optimization of autonomous mobile robots.
"Truly autonomous mobile robots have to solve the SLAM problem (i.e. simultaneous map building and pose estimation) to navigate in an unknown environment. This is a core problem to solve to reach real robot autonomy. Unfortunately, despite many solutions that exist, there is no robust and reliable solution for mobile platforms with limited resources. It’s always a trade-off between robustness, performance, and computational resource requirements. We research ways to improve SLAM quality without increasing computation/memory limits by using Dempster-Shafer probability and Transferable Belief Models."
Using a toy example (tinySLAM), a research team led by Kirill Krinkin showed how small enhancements can affect scan matching and occupancy tracking. Probability extensions could be one of the ways toward cheap smart spatial sensors.
Mikhail Yefremov, a graduate student of the Department of Software Engineering and Computer Applications, presented the results of research under the guidance of Ivan Kholod, Dean of the Faculty of Computer Science and Technology, in swarm robotics, a modern subdomain of robotics that studies fully autonomous robots without any central control or common knowledge base.
"One of the most common problems is the development process of such systems: at the moment there are no clear guides about designing and modeling of such systems, there is no standard of tools or software in the swarm robotics field. This work researches at swarm robotic systems from a software engineering point of view. The point of the research is how the software engineering approach can solve the problem of strong coupling between hardware platforms and swarm systems through guides and instruments for fast and high–quality swarm robotic application development."
The research team consisting of scientists from ETU "LETI" and the North-Caucasus Federal University, headed by Dmitry Kaplun, who was chair of the session on signal processing at the conference, and Pavel Lyakhov, the senior researcher of the Department of Automation and Control Processes, presented three reports, two of which focused on medical topics. The researchers told about the method for determining skin lesions from images using neural networks and high-performance hardware 3D medical imaging using wavelets in the residue number system, which allows increasing the performance of medical imaging devices by 1.57 – 2.70 times.
The third report of the team titled "Method of Oriented Contour Detection on Image Using Lorentz Function" discussed the area of digital image processing.
"The first distinguishing feature of the development is the ability to adjust the size of the filter mask to be able to vary the distance between the analyzed differences at the boundaries of the analyzed image areas. The second feature is the ability to pre-set the angle of rotation of the coordinate plane, which determines the orientation of the filter. In addition, the proposed filter has a minimum number of zones with different signs, which distinguishes it from the known Gabor filters. The proposed method can be used in various fields of digital image processing, but the most promising, in our opinion, is the use of the proposed filters in convolutional neural networks instead of convolutional layer neurons that are responsible for distinguishing features," Dmitry Kaplun highlighted.
Evgeniya Novikova, Associate Professor of the Department of Information Systems, presented a paper entitled "Exploration of the Anomalies in HVAC Data Using Image Similarity Assessment." The usage of the smart IoT devices has introduced the appearance of new hazardous scenarios of disruption, including possible attacks on heating, ventilation, and conditioning (HVAC) systems. The visualization-driven approaches may greatly facilitate the work of analysts when revealing anomalies in the dataset especially when there is no or little prior information about the HVAC system functioning.
"We developed an approach to the HVAC data analysis that enables the automated search of the days with similar HVAC functioning patterns based on image similarity analysis. The core idea of the approach is to form a graphical presentation of the HVAC data for each day and then assess their similarity by calculating the structural similarity index for each generated image. Days characterized by a high level of dissimilarity can be considered as anomalous ones requiring particular attention of an analyst. To test the approach proposed, we used the VAST MiniChallenge-2 2016 data set that contains logs from the HVAC system."
In total, researchers from four faculties of ETU "LETI" presented 34 papers on developments in the university's five priority R&D areas: Artificial Intelligence, Human-Machine Interface, Portable Medical Systems and Complexes, Advanced Wireless Technologies, Electrical Technologies and Power Engineering.
MECO Conference is a brand in the field of high technology. Papers presented at MECO are indexed in IEEE Xplore, Scopus, Web of Science, Google Scholar and all relevant databases. It has an SJR of 0.134 since 2012. Articles from the MECO conference were referenced in Google Scholar more than 5000 times, and some articles are published in Elsevier's SCIE journal Microprocessors and Microsystems with impact factor in WoS above 1.