Ernő Dávid, Dávid El-Saig, Zoltán Lehóczky and Gergely Gábor Barnaföldi (2020.01.01 - 2020.11.30)

Wigner Research Centre for Physics and Lombiq Technologies Ltd. cooperation

Publication: Implementing Hastlayer support for Xilinx SoC Zynq FPGA family

Abstract: Hastlayer by Lombiq Technologies allows software developers of the .NET platform to utilize FPGAs as compute accelerators. It converts standard .NET constructs into equivalent hardware implementations, automatically enhancing the performance while lowering the power consumption of suitable algorithms. Developers keep writing .NET programs as usual, no hardware design knowledge is required.

Hastlayer needs to support FPGA boards specifically, and formerly it only supported one that was suitable for testing and creating proof of concepts, but not for high performance computing scenarios. The recently finished work in collaboration with Wigner RCP was about making it support high-performance FPGAs of the Microsoft Catapult platform. Wigner's task was to create the FPGA-side hardware framework that hosts the automatically generated hardware cores created by Hastlayer.

The current project is similar, but about making Hastlayer support all SDAccel-compatible Xilinx FPGAs. This will open up a wide range of possibilities for Hastlayer-accelerated applications by unlocking the large product line of high-performance SDAccel-compatible FPGAs running on premise, as well as in multiple public cloud computing providers.

Károly Kubicskó, Ödön Farkas (2020.06.01 - 2020.11.30)

Eötvös Loránd University, Department of Organic Chemistry

Abstract: The metabolism of biogenic N,N-dimethyltryptamine (DMT) starts with its enzymatic oxidation to indole-3-N,N-dimethylethaniminium cation (imDMT+) by monoamine oxidase (MAO). Following this reaction, a non-enzymatic hydrolysis occurs, yielding indole-3-acetaldehyde (IAL) and dimethylamine (DMA). In our in silico study, using a hybrid DFT method at the M06-2X/aug-cc-pVDZ level, we examined the necessity of explicit water molecules to accurately model the hydrolysis. Our calculations revealed that despite an implicit continuum solvent model, one or two explicit solvent molecules should be added to obtain meaningful activation barriers. Thus, we demonstrate that the hydrolysis can occur in a human cell without enzymatic assistance, paving the way for deeper understanding of DMT metabolism.

Publication: Quantum chemical study of the hydrolysis of oxidized endogenous psychedelic N,N-dimethyltryptamine

Zoltán Zimborás, Dániel Nagy, András Németh, Gábor Németh (2020.05.22-2020.12.31)

Wigner Research Cetnre for Physics, Eötvös University Faculty of Science and Faculty of Informatics,Ericsson Research

Abstract: In recent years, both experimental quantum computers and quantum algorithms have been significantly improved. The advantage of quantum over classical computation is that if multiple qubits get entangled, the dimensionality of the Hilbert-space increases exponentially, and thus a quantum computer might be exponentially faster at solving certain problems than a classical one. Apart from this, the laws of quantum mechanics enable such computations and communication protocols, that are totally infeasible using classical computers. However, experimental quantum computers accessible at the moment contain only a few dozens of qubits, and those are quite noisy as well. Hence, when developing quantum algorithms, it is indispensable to use high performance classical computers to simulate these quantum algorithms in a reasonable time for a reasonable cost. Furthermore, an important research area is the investigation of hybrid quantum-classical algorithms, where the use of modern GPUs is necessary along with the quantum processors.

Forgács-Dajka Emese, Dobos László, Ballai István (2021.01.01-2021.09.30)

Eötvös Loránd Tudományegyetem, John Hopkins University és Sheffield University

Kivonat: A napaktivitás jelenségeinek megfigyelése nagy múltra tekint vissza, így lehetőségünk van hosszú időskálán történő vizsgálatok készítésére. Célunk a napfoltcsoportok területének időfejlődésének tanulmányozása egy 130 évet felölelő, különféle adatbázisokból álló, de konzisztens nagy mintában, a legmodernebb statisztikai módszerekkel. A gondosan leválogatott, de véletlenszerű minta esetében Bayes-modellt alkalmazunk, melynek segítségével a napfoltok teljes területére ferde normális függvényt illesztve meghatározzuk a folt növekedésének és bomlásának ütemét és az esetleges aszimmetriát. Ezen eljárás nagy előnye, hogy az olyan foltcsoportok esetén is megkaphatjuk a teljes élettartamra vonatkozó tulajdonságokat (pl. élettartam, fejlődési szakaszok különböző részei, mint felfutási, lefutási rész), melyekről nincsenek teljes adatsorok a megjelenéstől az eltűnésig. Ennek oka a megfigyelésben keresendő, azaz a Nap forgása miatt a foltcsoportok teljes életciklusa nem követhető végig minden esetben. Eredményeink robosztus módon alátámasztják azt a feltételezésünket, hogy a foltcsoportoknak két típusa van: (i) egy a megjelenést követően gyorsan kifejlődő, de lassabban lebomló, illetve (ii) egy lassabban felfutó, gyorsabban lecsengő. Emellett a modellparaméterek segítségével vizsgáltuk a napfoltcsoportok élettartamát, illetve az élettartam, maximális terület közötti összefüggéseket is.

Publikáció: Astronomy & Astrophysics, Volume 653, id.A50, 10 pp. DOI: 10.1051/0004-6361/202140731 arXiv: arXiv:2106.04917 Publisher link: https://www.aanda.org/articles/aa/pdf/2021/09/aa40731-21.pdf

József Kadlecsik (2019.12.01 - 2020.02.29)

Wigner RCP, SzHK

Publication: Laboratory study of water surface polygon vortices

Abstract: Although gravitational waves have been detected in 2016 with GW150914, but there are still many exotic cases to discover, e.g. binaries with Zoom-Whirl orbits, Super Massive Black Holes, etc. To examine such cases the Wigner RCP Gravitational group developed a software called CBwaves , that we would like to upgrade with the fourth terms of the post-Newtonian formalism and with its Hamiltonian formalism. Also we would like to enchance the runime of our code using parallelism.

Emese Forgács-Dajka, László Dobos, István Ballai (2021.03.01-2021.09.30)

Eötvös University, John Hopkins University, Sheffield University

Abstract: Abstract: Observing the phenomena of solar activity has a long history, so we have the opportunity to study on a long time scale. We aim to study the time dependence of sunspot group areas in a large sample composed of various databases spanning over 130 years, used state-of-the-art statistical methods. For a carefully selected but unbiased sample, we use Bayesian modelling to fit the temporal evolution of the combined umbral and penumbral area of spot groups with a skew-normal function to determine the existence of any asymmetry in spot growth or decay. The great advantage of this method is that we can obtain the properties for the whole lifetime (eg lifetime, different parts of the developmental stages, such as growth and decay part) for which there are no complete data sets from appearance to disappearance. The reason for this is to be found in the observation, i.e. due to the rotation of the Sun. Our results robustly support our hypothesis that there are two types of groups: (i) one that develops rapidly after appearance but decomposes more slowly, and (ii) one that grows more slowly and decays faster. In addition, we used the model parameters to examine the lifetime of the sunspot groups and the relationships between lifetime and maximum area.

Publication: Astronomy & Astrophysics, Volume 653, id.A50, 10 pp. DOI: 10.1051/0004-6361/202140731 arXiv: arXiv:2106.04917 Publisher link: https://www.aanda.org/articles/aa/pdf/2021/09/aa40731-21.pdf

David Legrady, Gabor Tolnai, Tamas Hajas, Előd Pázmán (2021.06.01 - 2022.04.30)

BME Institute of Nuclear Techniques

Publication: Full Core Pin-Level VVER-440 Simulation of a Rod Drop Experiment with the GPU-Based Monte Carlo Code GUARDYAN

Abstract: The GUARDYAN (GPU Assisted Reactor Dynamic Analysis, developed at BME Institute of Nuclear Techniques) Monte Carlo code directly follows the time evolution of the neutron field in a nuclear reactor. Contrary to the conventionally applied deterministic (i.e. non-Monte Carlo) or Monte-Carlo based techniques relying on quasistatic approximations modelling errors are minimal for GUARDYAN. For a fast evolving („hard”), localized transients even the magnitude of the modelling errors posed by conventional techniques can hardly be estimated, and experimental confirmation due to nuclear hazards is out of question. Therefore, simulations with GUARDYAN could be set as a gold standard for other computational methods. The project aims at the simulation of a rod ejection transient in a full-scale currently operational nuclear power plant type (VVER-440) using the code GUARDYAN.

Ernő Dávid, Dávid El-Saig and Zoltán Lehóczky (2019.08.31 - 2019.12.31)

Wigner RCP és Lombiq Technologies Ltd. cooperation

Abstract: Hastlayer by Lombiq Technologies allows software developers of the .NET platform to utilize FPGAs as compute accelerators. It converts standard .NET constructs into equivalent hardware implementations, automatically enhancing the performance while lowering the power consumption of suitable algorithms. Developers keep writing .NET programs as usual, no hardware design knowledge is required.

Hastlayer needs to support FPGA boards specifically, and formerly it only supported one that was suitable for testing and creating proof of concepts, but not for high performance computing scenarios. The work ongoing in collaboration with Wigner RCP is about making it support high-performance FPGAs of the Microsoft Catapult platform. Wigner's task is to create the FPGA-side hardware framework that hosts the automatically generated hardware cores created by Hastlayer.

Károly Kubicskó, Ödön Farkas

ELTE, The Department of Organic Chemistry

Publication: Quantum chemical (QM:MM) investigation of the mechanism of enzymatic reaction of tryptamine and N,N-dimethyltryptamine with monoamine oxidase A

Abstract: In our study we suggest a chemical mechanistic process between the biogen, strong hallucinogenic trace amine DMT and monoamine oxidase A (MAO-A) enzyme. We found that the hydride ion transfer from the alpha carbon atom of the amine is only favourable, when the flavin ring (in the enzyme) is protonated at a specific position. Our resuts are in accord with previous mechanistic studies in the research of metabolism of neurotransmitters by monoamine oxidase.

Hegedűs Tamás (2018.03.01-2019.08.31)

MTA-SE Molecular Biophysics Research Group, Hungarian Academy of Sciences

Publication: Quantitative comparison of ABC membrane protein type I exporter structures in a standardized way

Abstract: Cystic fibrosis is a fatal inherited monogenic recessive disease affecting several organs in our body (1:3000 prevalence in the Caucasian population). The disease is associated to the absence of functional CFTR (Cystic Fibrosis Transmembrane Conductance Regulator) chloride channel from the apical membrane of epithelial cells. Over 2,000 mutations are known covering all regions of the protein and the most frequent mutation is the deletion of F508 (ΔF508). This position is located in the N-terminal nucleotide binding domain (NBD1) resulting in the misfolding of this segment. We employ molecular dynamics simulations to detect differences between the dynamics of the wild type, ΔF508, and other mutant forms of NBD1. The simulations will be accelerated by GPU technologies that allow more and longer simulations than before, thus a more exhaustive characterization of the NBD1 conformational space. Our results will contribute to both drug development and understanding the effects of mutations at the atomic level.

Michał Bejger (2017.08.01-2019.08.31)

Nicolaus Copernicus Astronomical Center, Observatoire de Paris

Publication: Astronomical Distance Determination in the Space Age. Secondary distance indicators

Abstract: The aim of this project is to develop a production-ready version of the data-analysis pipeline to search for gravitational-wave signals from the network of Advanced Era LIGO and Virgo interferometric detectors. The algorithm developed by the Polish Virgo-POLGRAW group aims at finding almost-monochromatic gravitational-wave signals from rotating, non-axisymmetric, isolated neutron stars. The detection of such signals will open an exciting possibility of studying the physics of neutron-stars’ interiors, its elastic properties and structure of the crust. Joint project within the Hungarian high-performance computing experts and gravitational-wave experts will be beneficial for both sides and will initiate long-term collaboration in this field.

István Csabai PhD (2018.08.01-2018.12.31)

ELTE, The Department of Physics of Complex Systems

Publication: StePS: A Multi-GPU Cosmological N-body Code for Compactified Simulations

Abstract: Although the LCDM model has achieved remarkable success, however, in recent years the accuracy of the measurements has reached the limit where parameter estimates from various observations, such as the Hubble constant determined from both CMB and supernovae, are incompatible with it. Recently we have developed a model, based on N-body simulations, which is able to resolve this tension by taking better account of complex structure formation and without introducing dark energy. During the project we will develop a new type of \(N\)-body simulation algorithm ,,StePS" that overcomes limitations of current methods through mapping the infinite spatial extent of the universe onto a compact manifold. Specifically, we use stereographic projection onto the surface of a four dimensional sphere. The discretization of this surface leads to a systematic multi-resolution simulation with unprecedented dynamic range for given computational resources and perfect consistency with the Newtonian force law. Our approach retains the best features of multipole solvers and AMR simulations through a continuous, mathematically consistent refinement of scales toward the center of the simulations and constant angular resolution of distant fluctuations. The algorithm is ideal for GPUs, harnessing a recent cost effective numerical hardware revolution. A prototype of our algorithm has been successfully tested against GADGET, the early version of the code is open source, and available on GitHub, and the paper on the preliminary results R\'acz et al. (2018) has been submitted to MNRAS.

Richárd Forster (2017.01.01-2019.08.31)

CERN

Publication: Parallel Louvain Community Detection Optimized for GPUs

Abstract: Network analysis became a fundamental tool in understanding the structural and functional organization of the brain. The connectome of the cerebral cortex, the most complex part of the brain, is best known at the large scale, which, as a network, represents the connectivity between the different cortical areas or sub-regions. However, in reality brain areas and sub-regions are connected via numerous parallel pathways formed by populations of neurons residing within the areas, sub-regions. This mesoscale, or so called columnar network architecture of the cerebral cortex is not known, largely because of serious experimental limitations. However, graph theory provides some tools to approach the blueprint of the mesoscale cortical network.

The goal of the Neuroscience project is the application of graph theory and network analytic tools to explore the hidden structure of the large scale cortical network relevant to the mesoscale organization. To this end we study the network of the combinations of the incoming and outgoing edges of the different areas, which represent the interactions within the cortical network forming the interaction network. Considering cortical functioning, interactions or transfer between the inputs and the outputs is the basic operation of the areas. The interaction network is the two-hop representation of the network, which can be computed by graph derivations. Importantly, the derived network preserves the topological features of the original network but increases the size by orders of magnitude. The collaboration has been focusing on the exploration of the architecture of the interaction network. We aim to understand both the global organization of the interaction network as well as the role of the particular areas in regard to the specific pathways of the cortical signal flow.

collspotting.cern.ch (Accessible only with a CERN account)

Márton Vargyas (2016.09.01-2018.08.31)

git: TPCQA, HVsoft doi: 10.1016/j.nima.2018.06.084 arxiv: 1805.03234

Abstract: The task of the Budapest Advanced Quality Assurance (QA-A) Centre is to test and classify the Gas Electron Multiplier (GEM) foils, which would be an integral part of the upgrade of ALICE's Time Projection Chamber (TPC) detector. To classify the foils we use the known correlation between their hole size and electrical properties. We take high definition images of the foils in a clean room equipped with an X-Y-Z robot and telecentric lens (images of the foil's two sides can take up to 50GB), then we recognize these holes with a GPU-accelerated software, which identifies every hole. Then we decide the fate of the foil, either it would be built into the detector allowing continous readout, making it a 3D camera or we return it to the manufacturer.

Sándor Zsebők (2018.03.01-2019.08.31)

Department of Systematic Zoology and Ecology

Publication: Automatic bird song and syllable segmentation with an open-source deep-learning object detection method – a case study in the Collared Flycatcher (Ficedula albicollis)

Abstract: In the last two decades, many bioacoustic projects were conducted in the Department of Systematic Zoology and Ecology (Eötvös Loránd University). The works mainly related to the study of the structure and function of birdsong, the echolocation of bats and the development of passive acoustic methods to measure biological diversity in nature.
In our current research, we study the cultural evolution of the song of collared flycatcher based on large amount of acoustic data recorded in the nature. For that, first we have to find the songs in the recordings, then we have to segment the smallest units in the song (so called syllables), and at the end we have to cluster the syllables resulting a universal syllable library. This is a very time-consuming process that is intended to be done by computers. For that we would like to teach Deep Neural Networks. We possess hundreds of hours of recordings, thousands of songs and more than 150.000 manually segmented and clustered syllables. We would like to teach convolutional networks to detect the songs and syllables on raw recordings based on their spectrographic representations. This teaching process needs large capacity GPUs to be effectively done through parameter tuning. We would like to use the ready models for predicting songs and syllables in new recordings. Both this prediction and the clustering of syllables can be done on personal computers, so the application would only call for the training of the deep neural nets. The developing these models would facilitate all the projects that can be done only on large amount of collared flycatcher songs, and so it would open up new scientific directions to answer questions related to the evolution of animal acoustic communication.