Program

Son Hai Nguyen

neural solver, Helmholtz equation, graph neural network, transcranial ultrasound, PDE

Modelování a simulace Robotika a umělá inteligence

Neural solvers have been increasingly explored with the aim of replacing computationally expensive conventional numerical methods for solving PDEs. This work focuses on solving the time-independent Helmholtz equation for transcranial ultrasound therapy. Most of the popular methods for modeling physics systems are based on U-net. However, convolutional neural networks require the data to be sampled on a regular grid. In order to try to lift this restriction, we propose an iterative solver based on graph neural networks. Unlike Physics-informed neural networks, our model needs to be trained only once, and only a forward pass is required to obtain a new solution given input parameters. The model is trained using supervised learning, where the reference results are computed using the traditional solver k-Wave. Our results show the model’s unroll stability despite being trained with only 8 unroll iterations. Despite the model being trained on the data with a single source, it can predict wavefields with multiple sources and generalize to much larger computational domains. Our model can produce a prediction for sub-pixel points with higher accuracy than linear interpolation.

Dominik Boboš

Detection of re-occurring sequences in audio, Segmental dynamic time warping, Recurrence quantification analysis, Fuzzy string matching, Bottleneck features, Phoneme posteriors

Zpracování dat (obraz, zvuk, text apod.)

Recognition of pre-recorded messages in speech such as "This number is not reachable" is useful for any follow-up speech data mining. To investigate the identification of redundant information in audio, it is necessary to have a large amount of data with the exact phrases repeated multiple times. Such a set is generated by mixing pre-recorded messages into phone calls with variations in speed, volume and repetitions. The created system tackles “known messages” and “unknown messages” scenarios by using approaches like clustering or detection in chunks. Dynamic time warping, approximate string matching and recurrent quantification analysis are compared, and finally, all mentioned techniques are combined to obtain a precise and efficient system.

Tomáš Hladký

Simulator, DAG-based consensus, Blockchain, Optimizations, Payoff function, Transaction throughput

Bezpečnost

We study existing Directed Acyclic Graph (DAG) blockchain designs that propose to solve a blockchains throughput problem, especially protocols PHANTOM and its optimization GHOSTDAG. They utilize a Bitcoin protocol and propose a random transaction selection, resulting in increased transaction throughput. However, it has been proved by a simulation that actors that use the random transaction selection strategy have less profit than actors who do not follow the protocol and select transactions rationally (i.e., most profitable). That proof has been made on a small network of ten nodes with a circle topology. This article aims to extend, optimize, and automate an existing blockchain simulator. We implement a Bitcoin-like network topology with realistic block propagation latency. Furthermore, we optimize the simulator to run more simulations in parallel and faster, including automation tools that can create or edit input configurations, perform a combination of runs on multiple CPUs based on input parameters, and analyze profits and transaction collisions. Finally, we perform experiments to verify malicious actors' advantages in a Bitcoin-like network and create a payoff function to punish this behavior.

Barbora Šmahlíková

Büchi Automata, Elevator Automata, Büchi Complementation

Testování, analýza a verifikace

Büchi automata complementation is an important operation for decision procedures of some logics, proving termination of programs, or model checking of temporal properties. Due to the high space complexity of Büchi automata complementation, it is necessary to look for some optimizations reducing the size of generated state space for good performance in practice. In this paper, we identify elevator automata, a class of Büchi automata, which often occurs in practice. Thanks to their specific structure, we are able to reduce the bound on the maximum rank of states in each strongly connected component, which is one of the main causes of a state space blow-up in rank-based complementation. We compare these techniques, implemented as an extension of the tool Ranker, with other state-of-the-art tools and show that with these optimizations, rank-based complementation is competitive to other complementation approaches and can give better results on a large set of benchmarks.

Jindřich Šesták

ESP32, ESP-NOW, Mesh network, Mesh, Espressif, MicroPython, Asyncio

Počítačové sítě

The aim of this project is to implement a mesh network protocol on ESP32 microchips in MicroPython. It mainly focuses on the functioning of mesh in two modes, with connection to the Internet and without it. This thesis was ordered by Espressif company for improving and discovering new ways of mesh networking. % The solution of this mesh network uses two network protocols. First, the ESP-NOW protocol offers low power consumption and doesn't need any network connection. The second is the common WiFi protocol which is used for data transmission. WiFi links are formed between ESP32 nodes and one of the nodes can even be connected to the Internet and offer a connection to the whole mesh. % With full functionality, the mesh will be light weighted and will connect multiple nodes. It is possible to run user applications like light control on ESP32 boards on top of the mesh using WiFi. With WiFi, it is possible to transfer big amount of data for applications. The work is still in progress. % In this project, there are used new innovations to ensure the formation of a structure in the mesh. The problem of how to select a root node in an environment without the WiFi Access Point (Router, AP) is presented.

Petr Kohout

bioinformatika, proteinové inženýrstvı́, molekulárnı́ dynamika, tunely, dokovánı́

Bioinformatika

Tato práce se zabývá analýzou proteinových struktur. Cílem je vytvořit Caver Web 2.0 - novou verzi webové aplikace, která začlení další vědecké nástroje a uživatelům umožní projít komplikovaný pracovní protokol za poskytnutí relevantních výsledků bez nutnosti hlubší znalosti integrovaných nástrojů. Vše bude zprostředkováno prostřednictvı́m jednoduchého a interaktivnı́ho uživatelského rozhranı́. Aplikace rozšiřuje původnı́ aplikaci Caver Web 1.0 o nové vlastnosti. Caver Web 1.0 je webový server vhodný pro identifikaci proteinových tunelů a kanálů, pro které umožňuje spustit analýzy transportu ligandů. Program se vyznačuje intuitivnı́m a uživatelsky přı́větivým rozhranı́m s minimem požadovaných vstupů od uživatele. Server je vhodný i pro výzkumnı́ky bez pokročilých bioinformatických nebo technických znalostı́. Jeho současná verze je ve vědecké komunitě dobře zavedená a velmi využı́vaná (35 000 dokončených výpočtů během dvou let provozu). Nejvýznamnějšı́m omezenı́m současné verze je možnost analyzovat pouze statickou strukturu, což často poskytuje neúplný biologický obraz. Proto jsme se rozhodli nástroj rozšı́řit o výpočet molekulárnı́ch dynamik, které poskytnou ucelený obraz na proměny proteinových struktur. Touto funkcı́ se Caver Web 2.0 stane prvnı́m webovým nástrojem, který bude poskytovat analýzu tunelů bez nutnosti ručnı́ch výpočtů molekulárnı́ch dynamik. Grafické uživatelské rozhranı́ bude navrženo speciálně pro vědeckou komunitu s podporou jednoduchého exportu. Nástroj bude zdarma k dispozici celé vědecké komunitě.

Samuel Repka

Photometric stereo, Scanning electron microscope, 4-segment BSE detector

Zpracování dat (obraz, zvuk, text apod.)

While scanning electron microscopes (SEM) provide an extraordinary spectrum of information about the specimens, they do not answer a few fundamental questions. How high? How deep? This paper aims to explore possibilities of topography reconstruction of microscopic samples, as well as to attempt to solve the task using tools already available on conventional scanning electron microscopes. The proposed solution uses images from a four-segment backscattered electrons detector as an input to the photometric stereo algorithm. This algorithm exploits the fact, that the brightness of the image point is dependent on the inclination of the sample surface. Reflectance maps are used to estimate the inclination in each pixel, creating a map of normal vectors. The map is then used for topography reconstruction. A novel technique for reflectance map estimation is proposed. This method is applied to tin samples to remove the sample’s atomic number effects. Solution evaluation was not yet performed. The fact that all data are acquired simultaneously allows for fast reconstruction. Usage of already available and widespread tools eliminate a need for specialized equipment such as Atomic Force Microscopes.