BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//CERN//INDICO//EN
BEGIN:VEVENT
SUMMARY:Massively Parallel Tensor Network Algorithms
DTSTART;VALUE=DATE-TIME:20220621T142000Z
DTEND;VALUE=DATE-TIME:20220621T144000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3185@indico.wigner.hu
DESCRIPTION:Speakers: Andor Menczer (Eötvös Loránd University (ELTE-IK)
)\nAs the design and mass manufacturing of efficient quantum computers are
still subject of intense research\, the numerical simulations of quantum
systems still rely on classical computation. In this case however the comp
lexity and resource requirements of such algorithms scale exponentially re
lative to the system size\, thus making bigger simulations problematic or
even impossible to run.\n\nOur approach focuses on the development of mass
ively parallel algorithms that are not only highly scalable and ideal to u
se in an HPC environment\, but by building on the foundation of theoretica
l physics and applied mathematics the number of required arithmetic calcul
ations could be reduced by multiple magnitudes. As a result the exponentia
l time cost of the simulations has collapsed into polynomial complexity.\n
\nThe research program puts an emphasis on one of the subclasses of tensor
network state algorithms called density matrix renormalization group\, or
DMRG for short. In such cases large-scale tensor operations can be substi
tuted with multi-million vector and matrix operations\, of which many can
be executed independently of one another. Through the exploitation of thes
e (in)dependencies arithmetic operations can be reordered and put into mul
tiple tiers of groups corresponding to specific software and hardware laye
rs ranging from low level CPU and GPU based SIMD execution to high level H
PC scheduling. Thanks to the fact that for every tier we can execute all o
perations contained within the same group independently of all other arith
metics residing outside the group\, mass scale parallelism can be achieved
at every tier of our multi-tiered grouping. The resulting parallelization
is the product of each tier's own massive parallelization\, thus with sui
table hardware infrastructure exascale computing in the near future might
become a reality for DMRG based quantum simulations.\n\nhttps://indico.wig
ner.hu/event/1393/contributions/3185/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3185/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Studying hadronization with Machine Learning techniques
DTSTART;VALUE=DATE-TIME:20220621T090000Z
DTEND;VALUE=DATE-TIME:20220621T092000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3160@indico.wigner.hu
DESCRIPTION:Speakers: Gabor Biro (MTA Wigner FK)\nHadronization is a non-p
erturbative process\, which theoretical description can not be deduced fro
m first principles. Modeling hadron formation\, requires several assumptio
ns and various phenomenological approaches. Utilizing state-of-the-art Com
puter Vision and Deep Learning algorithms\, it is eventually possible to t
rain neural networks to learn non-linear and non-perturbative features of
the physical processes.\n\nHere\, I would like to present the results of t
wo ResNet networks\, by investigating global and kinematical quantities\,
indeed jet- and event-shape variables. The widely used Lund string fragmen
tation model is applied as a baseline in √s=7 TeV proton-proton collisio
ns to predict the most relevant observables at further LHC energies. Non-l
iear QCD scaling properties were also identified and validated by experime
ntal data.\n\n[1] G. Bíró\, B. Tankó-Bartalis\, G.G. Barnaföldi\; arXi
v:2111.15655\n\nhttps://indico.wigner.hu/event/1393/contributions/3160/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3160/
END:VEVENT
BEGIN:VEVENT
SUMMARY:The resonant structure of the trans-Neptunian space
DTSTART;VALUE=DATE-TIME:20220621T140000Z
DTEND;VALUE=DATE-TIME:20220621T142000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3153@indico.wigner.hu
DESCRIPTION:Speakers: Emese Kővári (Department of Astronomy\, Institute
of Geography and Earth Sciences\, Eötvös Loránd University\, H-1117 Bud
apest\, Pázmány Péter sétány 1/A\, Hungary\; Centre for Astrophysics
and Space Science\, Eötvös Loránd University\, H-1117 Budapest\, Pázm
ány Péter sétány 1/A\, Hungary\; Wigner Research Centre for Physics\,
P.O. Box 49\, Budapest H-1525\, Hungary)\nThe outer realm of the Solar Sys
tem\, known either as the trans-Neptunian space or the Kuiper belt\, is of
great interest among celestial mechanical studies. Its dynamical structur
e is shaped to a large extent by the mean-motion resonances (MMRs) occurri
ng between the trans-Neptunian objects (TNOs) and (mainly) the Neptune.\nI
n a recent research\, we carried out a large-scale survey of the TNOs\, wi
th a sample containing more than 4100 small bodies. By applying the FAIR m
ethod (Forgács-Dajka\, Sándor\, and Érdi\, 2018)\, we identified the mo
st important MMRs\, and distinguished between the so-called short- and lon
g-term resonances: TNOs in the former category are only temporarily captur
ed in a given MMR\, while those of the latter remain resonant throughout t
he (sufficiently long) integration time. We explored the dynamical propert
ies of such intriguing MMRs through the quantification of the chaotic diff
usion. For this purpose\, we adopted both classical methods (as the one e.
g. based on the computation of the time evolution of the variance of an ac
tion variable) and a more recent one (based on the computation of the time
derivative of the Shannon entropy).\nApart from investigating the individ
ual TNOs of our sample\, we constructed dynamical maps of fictitious test
particles\, too. This approach enabled us to analyze the structure of the
phase space in the vicinity of the resonances.\nOur findings indicate that
a notable percentage of the examined TNOs are engaged in MMRs with Neptun
e\, and that however chaotic the phase space appears to be\, the diffusion
timescales are remarkably long.\nAs for the technical realization of the
research\, we adopted a barycentric model of the Solar System - containing
the four giant planets and either a massless TNO or a test particle -\, a
nd integrated the equations of motion on a timescale of million years. Our
codes were optimized for GPU computations in order to deal with the signi
ficant computational costs of integrating several hundreds of thousands of
initial conditions (i.e. test particles).\n\nhttps://indico.wigner.hu/eve
nt/1393/contributions/3153/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3153/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Machine learning Hadron Spectral Functions in Lattice QCD
DTSTART;VALUE=DATE-TIME:20220621T094000Z
DTEND;VALUE=DATE-TIME:20220621T100000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3187@indico.wigner.hu
DESCRIPTION:Speakers: Feiyi Liu (Eötvös Loránd University & Central Chi
na Normal University)\nWe study the inverse problem of reconstructing spec
tral functions from Euclidean correlation functions via machine learning.
We propose a novel neural network\, SVAE\, which is based on the variation
al autoencoder (VAE) and can be naturally applied to the inverse problem.
The prominent feature of the SVAE is that a Shannon-Jaynes entropy term ha
ving the ground truth values of spectral functions as prior information is
included in the loss function to be minimized. We train the network with
general spectral functions produced from a Gaussian mixture model. As a te
st\, we use correlators generated from four different types of physically
motivated spectral functions made of one resonance peak\, a continuum term
and perturbative spectral function obtained using non-relativistic QCD. F
rom the mock data test we find that the SVAE in most cases is comparable t
o the maximum entropy method (MEM) in the quality of reconstructing spectr
al functions and even outperforms the MEM in the case where the spectral f
unction has sharp peaks with insufficient number of data points in the cor
relator. By applying to temporal correlation functions of charmonium in th
e pseudoscalar channel obtained in the quenched lattice QCD at 0.75 $T_c$
on $128^3\\times96$ lattices and $1.5$ $T_c$ on $128^3\\times48$ lattices\
, we find that the resonance peak of $\\eta_c$ extracted from both the SVA
E and MEM has a substantial dependence on the number of points in the temp
oral direction ($N_\\tau$) adopted in the lattice simulation and $N_\\tau$
larger than 48 is needed to resolve the fate of $\\eta_c$ at 1.5 $T_c$.\n
\nhttps://indico.wigner.hu/event/1393/contributions/3187/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3187/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Online data processing with GPUs in ALICE during LHC Run 3
DTSTART;VALUE=DATE-TIME:20220620T120000Z
DTEND;VALUE=DATE-TIME:20220620T124000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3186@indico.wigner.hu
DESCRIPTION:Speakers: David Rohr (CERN)\nThe ALICE experiment has undergon
e a major upgrade for LHC Run 3 and will record 50 times more heavy ion co
llisions than before.\nThe new computing scheme for Run 3 replaces the tra
ditionally separate online and offline frameworks by a unified one.\nProce
ssing will happen in two phases.\nDuring data taking\, a synchronous proce
ssing phase performs data compression\, calibration\, and quality control
on the online computing farm.\nThe output is stored on an on-site disk buf
fer.\nWhen there is no beam in the LHC\, the same computing farm is used f
or the asynchronous reprocessing of the data which yields the final recons
truction output.\nALICE will employ neither hardware nor software triggers
for Pb-Pb data taking but instead store all collisions in compressed form
.\nThis requires full online processing of all recorded data\, which is a
major change compared to a traditional online systems\, which sees only th
e data selected by a hardware trigger.\nTraditional CPUs are unable to cop
e with the huge data rate and processing demands of the synchronous phase\
, therefore ALICE employs GPUs to speed up the processing.\nSince the onli
ne computing farm performs a part of the asynchronous processing\, ALICE p
lans to use the GPUs also for this second phase when there is no beam in t
he LHC.\nThe primary goal for the commissioning in 2021 and 2022 was to ma
ke those reconstruction steps required for the online phase run on the GPU
efficiently.\nThe development is now shifting towards moving more computi
ng-intensive steps of the asynchronous reconstruciton to the GPU as well.\
nThe talk will detail the ALICE Run 3 computing scheme\, and outline the h
ardware architecture and software design for synchronous and asynchronous
processing.\n\nhttps://indico.wigner.hu/event/1393/contributions/3186/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3186/
END:VEVENT
BEGIN:VEVENT
SUMMARY:What makes us humans: Differences in the critical dynamics underly
ing the human and fruit-fly connectome
DTSTART;VALUE=DATE-TIME:20220620T101000Z
DTEND;VALUE=DATE-TIME:20220620T103000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3150@indico.wigner.hu
DESCRIPTION:Speakers: Geza Odor (EK-MFA)\nPrevious simulation studies on h
uman connectomes [1] suggested\, that critical dynamics emerge subcrititca
lly in the so called Griffiths Phases. This is the consequence of the stro
ng heterogeneity of the graphs. Now we investigate this on the largest ava
ilable brain network\, the $21.662$ node fruit-fly connectome\, using the
Kuramoto synchronizationmodel. As this graph is less heterogeneous\, lacki
ng modular structure and exhibit high topological dimension\, we expect a
difference from the previous results. Indeed\, the synchronization transit
ion is mean-field like\, and the width of the transition region is larger
than in random graphs\, but much smaller than as for the KKI-18 human conn
ectome. This demonstrates the effect of modular structure and dimension on
the dynamics\, providing a basis for better understanding the complex cri
tical dynamics of humans [2].\nI show some numerical results obtained by t
he Kuramoto-GPU code developed for ODE solution of synchronization phenome
na.\n\n[1] G. Odor and J. Kelling\, Critical synchronization dynamics of t
he Kuramoto model on connectome and small world graphs\, Scientic Reports
9 (2019) 19621.\n\n[2] Geza Odor\, Gustavo Deco and Jeffrey Kelling\nDiffe
rences in the critical dynamics underlying the human and fruit-fly \nconne
ctome\, Phys. Rev. Res. 4 (2021) 023057.\n\nhttps://indico.wigner.hu/event
/1393/contributions/3150/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3150/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Representation learning in Artificial Intelligence
DTSTART;VALUE=DATE-TIME:20220620T132000Z
DTEND;VALUE=DATE-TIME:20220620T134000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3177@indico.wigner.hu
DESCRIPTION:Speakers: Antal Jakovac (Wigner RCP)\nIn Artificial Intelligen
ce (AI) the success of learning depends crucially on the way we represent
the input data. In the talk we overview the criteria of an ideal represent
ation\, and associate an entropy formula to them. We also show how these r
epresentations work in case of mechanical motion reconstruction from data.
\n\nhttps://indico.wigner.hu/event/1393/contributions/3177/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3177/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Application of high-performance computing for bubble simulations i
n sonochemistry
DTSTART;VALUE=DATE-TIME:20220621T150000Z
DTEND;VALUE=DATE-TIME:20220621T152000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3167@indico.wigner.hu
DESCRIPTION:Speakers: Dániel Nagy (Budapest University of Technology and
Economics\, Department of Hydrodynamic systems)\nThe objective of sonochem
istry is to increase the yield of chemical processes in a fluid with ultra
sound excitation. It is based on a special case of cavitation called acous
tic cavitation. Because of the ultrasound excitation\, several bubbles and
bubble-clouds can be formed in a liquid. During the radial oscillation of
the bubbles\, their compression can be so large that the internal tempera
ture can reach several thousands of Kelvins inducing chemical reactions. T
he importance of sonochemistry is in its potential applications\, e.g.\, n
ano-metal particle production\, organic synthesis\, or water purification.
\nUnderstanding the behaviour of a single bubble in an acoustic field is
an important topic of sonochemistry\, with many open questions. The presen
tation focuses on the break-up mechanism of a bubble into several smaller
bubbles. To directly observe the behaviour of a single bubble\, a computat
ional approach is used (CFD). This requires a multiphase model (gas and li
quid). Another difficulty is the differences in spatial scales: the size o
f a bubble in sonochemistry is usually a few micrometres\, while the wavel
ength of the used ultrasound is a few millimetres. An appropriate spatial
resolution of the problem requires a highly resolved adaptive mesh with mi
llions of cells. Furthermore\, the problem must be solved in time with an
appropriately small step size to correctly simulate the propagation of aco
ustic waves around the bubble. **Due to the high spatial and temporal reso
lution\, the solution must be parallelized\, and supercomputers must be us
ed to reduce the runtime of the simulations.**\nTo simulate a single bubbl
e\, the open-source program package called ***ALPACA*** is used\, which is
capable of simulating compressible multiphase flows in 2D or 3D. A multir
esolution algorithm is employed to automatically adapt the numerical mesh
during the solution process\; thus\, it is suitable for bubble simulations
. Moreover\, *ALPACA* is designed to be run on supercomputers. During the
research\, the *SUPERMUC-NG* supercomputer in Germany is used in collabora
tion with the *Nanoshock* research group from the *Technical University of
Munich*. \nIn this presentation\, **we demonstrate the basics of a bubble
simulation and analyse the strong scaling of such simulations for up to 3
00 CPU cores.** Based on the scaling analysis\, an appropriate configurati
on can be found for efficient simulations with which several different par
ameters (e.g.\, bubble radius\, acoustic excitation frequency) can be test
ed. Finally\, the results of the bubble simulations are discussed.\n\nhttp
s://indico.wigner.hu/event/1393/contributions/3167/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3167/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Piquasso\, a comprehensive framework for optical quantum computer
programming and simulation
DTSTART;VALUE=DATE-TIME:20220621T070000Z
DTEND;VALUE=DATE-TIME:20220621T073000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3184@indico.wigner.hu
DESCRIPTION:Speakers: Zoltan Zimboras (Wigner RCP)\nIn this talk\, we intr
oduce Piquasso\, a full-stack open source platform for Photonic Quantum Co
mputing built using Python and C++. Piquasso enables users to perform effi
cient Quantum Computing using continuous variables\, which could be used f
or designing photonic circuits for simulation and machine learning purpose
s.\n\nhttps://indico.wigner.hu/event/1393/contributions/3184/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3184/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Closing
DTSTART;VALUE=DATE-TIME:20220621T152000Z
DTEND;VALUE=DATE-TIME:20220621T154000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3182@indico.wigner.hu
DESCRIPTION:Speakers: Gergely Barnafoldi (Wigner RCP RMI of the Hungarian
Academy of Sciences)\nhttps://indico.wigner.hu/event/1393/contributions/31
82/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3182/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Machine learning methods for Schlieren imaging of a plasma channel
in tenuous atomic vapor
DTSTART;VALUE=DATE-TIME:20220620T140000Z
DTEND;VALUE=DATE-TIME:20220620T142000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3151@indico.wigner.hu
DESCRIPTION:Speakers: Mihály András Pocsai (Wigner RCP)\nInventing and f
ine-tuning laser and plasma based electron accelerators is a hot topic of
contemporary physics\, either considering experimental\, theoretical or ap
plied physics. One of the most prominent experiments in this field is the
CERN-AWAKE experiment [1]. In this experiment\, electrons are accelerated
by the wakefields generated by a series of proton microbunches in a 10-met
er-long rubidium plasma channel. The series of proton microbunches is gene
rated via the self-modulation instability: first\, the proton beam\, obtai
ned from the SPS experiment\, enters the plasma channel\, then the head of
the proton beam generates plasma wakes that split the proton beam into a
series of microbunches with a length of a few tens of micrometers each.\nT
he plasma itself is generated via photoionisation of rubidium vapour with
an ionising laser pulse with $780 \\\, \\mathrm{nm}$ wavelength\, $120 \\\
, \\mathrm{fs}$ pulse duration and $450 \\\, \\mathrm{mJ}$ pulse energy [2
].\n\nThe spatial extent of the plasma channel that is generated by the io
nizing laser pulse can be investigated using a Schlieren imaging setup. To
obtain parameters for the extent of the plasma channel\, we assume the pl
asma density distribution to be of the form:\n\\begin{equation}\n \\mathca
l{N}_{plasma} = \\left\\{ \n \\begin{aligned}\n &\\mathcal{N}_0 P_{max}\,
\\mathrm{~if~} r\\leq r_0\\\\\n &\\mathcal{N}_0 P_{max}\\exp\\left(-\\frac
{(r-r_0)^2}{t_0^2}\\right) \, \\mathrm{~if~} r>r_0\n \\end{aligned}\\righ
t.\n\\end{equation}\nwith $\\mathcal{N}_0$ being the vapour density\, $P_{
max}$ the maximum of the photoionisation probability\, i.e.~ the value mea
sured in the center of the plasma channel\, $r=\\sqrt{(y-y_0)^2+z^2}$ the
distance from the center of the plasma channel that is located at $(y\,z)
=(y_0\,0)$ and $r_{0}$ the radius of the plasma channel. $t_{0}$ character
ises the width of the region where the photoionisation probability rises f
rom $0$ to $P_{max}$. The output of the imaging setup can be calculated in
a straightforward way with any given plasma density distribution. The rev
erse is not true\, however\, as the measured image depends on the plasma c
hannel parameters $P_{max}\, y_0\, r_0\, t_0$ in a complicated way. The ta
sk of inverting the problem\, that is\, to determine the plasma parameters
from the calculated image can be attempted by using machine learning meth
ods. Below we shortly summarize the key features of our approach and our r
ecent results. We invite the Reader to look at our paper\, currently avail
able on arXiv\, for a detailed description [3].\n\nUsing computer simulati
ons\, a sufficient amount of good quality learning data can be generated w
ith low computational costs. Machine learning methods have the capability
of determining the parameters of the plasma density distribution\, shown i
n the Equationgiven above. We applied different deep neural network archit
ectures to achieve the goal\, and will present three models that produce t
he best predictions. Our results show that using a machine learning approa
ch\, the plasma parameters can be determined with high accuracy\, regardle
ss of the background noise. We also compared the predicted Schlieren signa
ls with the reference signals and experienced that our neural networks pre
dicted the signals themselves accurately\, with only a few percents of mea
n amplitude error and phase error. The calculated probability distribution
s of these errors also confirm the high accuracy of the predictions. Furth
ermore\, we tested how sensitive our networks are to the uncertainty of th
e vapour density and the probe laser beam intensity. We found if the actua
l vapor density or the probe laser intensity differs not more than $\\sim
2.5 \\\, \\%$ from the reference value\, i.e.~the value for which our netw
orks have been trained\, the accuracy of the predictions remains acceptabl
e. This suggests that our approach is a reliable\, robust method\, with po
ssibly better performance than other\, classical methods\, and is suitable
for the automated evaluation of experimental data.\n\n**References**\n\n
1. E. Gschwendtner\, et al.\, AWAKE\, the Advanced Proton Driven Plasma Wa
keeld Acceleration experiment at CERN\, Nuclear Instruments and Methods in
Physics Research Section A: Accelerators\, Spectrometers\, Detectors and
Associated Equipment 829 (2016) 76-82. 2nd European Advanced Accelerator C
oncepts Workshop - EAAC 2015.\n 2. E. Adli\, A. Ahuja\, O. Apsimon\, R. Ap
simon\, A.-M. Bachmann\, D. Barrientos\, F. Batsch\, J. Bauche\, V. B. Ols
en\, M. Bernardini\, et al.\, Acceleration of electrons in the plasma wake
field of a proton bunch\, Nature 561 (2018) 363.\n 3. G. Bíró\, M. A. Po
csai\, I. F. Barna\, J. T. Moody\, G. Demeter\, Machine learning methods f
or schlieren imaging of a plasma channel in tenuous atomic vapor\, 2022. U
RL: https://arxiv.org/abs/2205.12731 . doi:10.48550/arxiv.2205.12731.\n\nh
ttps://indico.wigner.hu/event/1393/contributions/3151/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3151/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Strategies for multi-GPU PIC/MCC plasma simulation implementation
on pre-exascale supercomputers
DTSTART;VALUE=DATE-TIME:20220620T142000Z
DTEND;VALUE=DATE-TIME:20220620T144000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3149@indico.wigner.hu
DESCRIPTION:Speakers: Zoltan Juhasz (University of Pannonia)\nIn this talk
we describe\, compare and evaluate various implementation strategies that
can be used to implement massively parallel Particle-in-Cell / Monte Carl
o collision low-pressure plasma simulations. Building on our earlier singl
e-GPU 1D and 2D plasma implementations that demonstrated two orders of mag
nitude speedup\, our goal is now to utilise the thousands of GPUs found in
pre-exascale supercomputers such as MARCONI 100\, SUMMIT or LEONARDO. A k
ey performance bottleneck in these distributed memory GPU systems is commu
nication cost. Traditionally\, these systems are programmed in a hybrid pa
rallel fashion using a combination of CUDA\, OpenMP and MPI for controllin
g and coordinating different levels of parallelism that results in a compl
ex and architecture dependent simulation code achieving -- at best -- weak
scaling only. We demonstrate these traditional multi-GPU programming stra
tegies in the context of our 1D plasma simulation program. We will illustr
ate the limitation of these approaches and their inability to hide inter-G
PU communication efficiently. Then\, we overview two alternative approache
s\, NCCL and NVSHMEM\, that provide device-side\, kernel-initiated communi
cation operations that provide a more GPU-friendly programming model and i
mproved communication hiding capabilities. Our work is still in progress b
ut we will show preliminary results that can demonstrate the design and im
plementation challenges of large multi-GPU programs.\n\nhttps://indico.wig
ner.hu/event/1393/contributions/3149/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3149/
END:VEVENT
BEGIN:VEVENT
SUMMARY:pCT Image Reconstruction -- A Huge Linear Problem
DTSTART;VALUE=DATE-TIME:20220620T130000Z
DTEND;VALUE=DATE-TIME:20220620T132000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3171@indico.wigner.hu
DESCRIPTION:Speakers: Akos Sudar (MTA Wigner FK)\nModern proton Computed T
omography (pCT) images are usually reconstructed by algebraic reconstructi
on techniques (ART). The Kaczmarz-method and its variations are among the
most widely used methods\, which are iterative solution techniques for lin
ear problems with sparse matrices. It is an interesting question whether s
tatistically-motivated iterations\, which have been successfully used for
emission tomography\, can be applied to reconstruct the novel technology o
f pCT images as well. \n\nIn my research\, I developed a method for pCT im
age reconstruction\, based on the Richardson–Lucy deconvolution. It trea
ts the problem as a statistically-motivated fixed-point iteration. I imple
mented this algorithm as a parallel code to GPU\, with spline-based trajec
tory calculation and on-the-fly system matrix generation. My results prese
nted that the method works well\, and it can be successfully applied in pC
T applications\, such as in the detector R&D of the Bergen pCT Collaborati
on.\n\nhttps://indico.wigner.hu/event/1393/contributions/3171/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3171/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Numerical Simulation of Mirages Above Water Bodies
DTSTART;VALUE=DATE-TIME:20220621T123000Z
DTEND;VALUE=DATE-TIME:20220621T125000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3180@indico.wigner.hu
DESCRIPTION:Speakers: Balázs Bámer (Wigner Research Center for Physics)\
, Anna Horváth (Wigner Research Center for Physics)\nWhen light travels t
hrough a medium with a changing refractive index\, it gets bent towards it
s higher values. This can produce upside down "mirror" images of a scenery
in deserts\, over heated roads or above bodies of water. We built a model
for the temperature profile based on measurement data\, from which we cal
culated the refractive index of air using another model. We implemented a
computer program\, which simulates mirages above water using the method of
ray tracing. Rays of light are traced by solving the eikonal equation usi
ng different Runge-Kutta methods. We developed a simple extension of the R
unge-Kutta method to efficiently check ray intersection with an arbitrary
geometry. Given a picture\, a physical setup and a value for both the temp
erature of the water body and that of the ambient air sufficiently far fro
m the surface (where it can be considered constant)\, our program can real
istically reproduce images of photographed mirages.\n\nhttps://indico.wign
er.hu/event/1393/contributions/3180/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3180/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Machine learning based estimator for elliptic flow in heavy-ion co
llisions
DTSTART;VALUE=DATE-TIME:20220620T124000Z
DTEND;VALUE=DATE-TIME:20220620T130000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3162@indico.wigner.hu
DESCRIPTION:Speakers: Suraj Prasad (Indian Institute of Technology Indore
(IN))\nUsing the kinematic information of the final state particles produ
ced in heavy-ion collisions at relativistic energies\, one tries to probe
the properties of the very hot and dense medium formed just after the coll
ision. There have been different probes to study the physics associated wi
th such a medium\, and one of them is the elliptic flow ($v_2$). In this s
tudy\, we have employed a deep neural network (DNN) based estimator in the
machine learning framework to estimate $v_2$\, using the particle kinemat
ic information as the input. The DNN model is trained with Pb-Pb collision
s at $\\sqrt{s_{NN}}$ = 5.02 TeV minimum bias data\, simulated with AMPT.
The trained model is also evaluated for Pb-Pb collisions at $\\sqrt{s_{NN}
}$ = 2.76\, 5.02 TeV and Au-Au collisions at $\\sqrt{s_{NN}}$ = 200 GeV\,
and is compared with ALICE experimental results. The proposed DNN model pr
eserves the centrality\, and transverse momentum dependence of the flow co
efficient. It is also found to be quite sturdy when subjected to simulated
data with the uncorrelated noise as the prediction accuracy of the DNN mo
del remains intact upto a reasonable extent. Such an estimator is yet to b
e tested with the experimental inputs along with detector level correlatio
ns in future with ALICE.\n\nhttps://indico.wigner.hu/event/1393/contributi
ons/3162/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3162/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Full Core Pin-Level VVER-440 Simulation of a Rod Drop Experiment w
ith the GPU-Based Monte Carlo Code~GUARDYAN
DTSTART;VALUE=DATE-TIME:20220621T131000Z
DTEND;VALUE=DATE-TIME:20220621T134000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3170@indico.wigner.hu
DESCRIPTION:Speakers: David Legrady (Dr.)\nTargeting ultimate fidelity rea
ctor physics calculations the Dynamic Monte Carlo (DMC) method simulates r
eactor transients without resorting to static or quasistatic approximation
s. Due to the capability to harness the computing power of Graphics Proces
sing Units\, the GUARDYAN (GpU Assisted Reactor DYnamic ANalysis) code has
been recently upscaled to perform pin-by-pin simulations of power plant s
cale systems as demonstrated in this contribution. A recent rod drop exper
iment at a VVER-440/213 (vodo-vodyanoi enyergeticheskiy reaktor) type powe
r plant at Paks NPP\, Hungary\, was considered and signals of ex-core dete
ctors placed at three different positions were simulated successfully by G
UARDYAN taking realistic fuel loading\, including burn-up data into accoun
t. Results were also compared to the time-dependent Paks NPP in-house noda
l diffusion code VERETINA (VERONA: VVER Online Analysis and RETINA: Reacto
r Thermo-hydraulics Interactive). Analysis is given of the temporal and sp
atial variance distribution of GUARDYAN fuel pin node-wise power estimates
. We can conclude that full core\, pin-wise DMC power plant simulations us
ing realistic isotope concentrations are feasible in reasonable computing
times down to 1--2\\% error of ex-core detector signals using current NVID
IA A100 GPU High Performance Computing architectures\, thereby demonstrati
ng a technological breakthrough.\n\nhttps://indico.wigner.hu/event/1393/co
ntributions/3170/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3170/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Critical synchronization dynamics on power grids
DTSTART;VALUE=DATE-TIME:20220621T144000Z
DTEND;VALUE=DATE-TIME:20220621T150000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3164@indico.wigner.hu
DESCRIPTION:Speakers: Shengfeng Deng (Institute of Technical Physics and M
aterials Science\, Centre for Energy Research)\nDynamical simulation of th
e cascade failures on the EU and USA high-voltage power grids has been don
e via solving the second-order Kuramoto equation. We show that synchroniza
tion transition happens by increasing the global coupling parameter $K$ wi
th metastable states depending on the initial conditions so that hysteresi
s loops occur. We provide analytic results for the time dependence of freq
uency spread in the large $K$ approximation and by comparing it with numer
ics of $d=2\,3$ lattices\, we find agreement in the case of ordered initia
l conditions. However\, different power-law (PL) tails occur\, when the fl
uctuations are strong. After thermalizing the systems we allow a single li
ne cut failure and follow the subsequent overloads with respect to thresho
ld values $T$. The PDFs $p(N_f)$ of the cascade failures exhibit PL tails
near the synchronization transition point $K_c$. Near $K_c$ the exponents
of the PL-s for the US power grid vary with $T$ as $1.4 \\le \\tau \\le 2.
1$\, in agreement with the empirical blackout statistics\, while on the EU
power grid we find somewhat steeper PL-s characterized by $1.4 \\le \\tau
\\le 2.4$. Below $K_c$ we find signatures of $T$-dependent PL-s\, caused
by frustrated synchronization\, reminiscent of Griffiths effects. Here we
also observe stability growth following the blackout cascades\, similar to
intentional islanding\, but for $K > K_c$ this does not happen. For $T <
T_c$\, bumps appear in the PDFs with large mean values\, known as ``dragon
king'' blackout events. We also analyze the delaying/stabilizing effects
of instantaneous feedback or increased dissipation and show how local sync
hronization behaves on geographic maps.\n\nhttps://indico.wigner.hu/event/
1393/contributions/3164/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3164/
END:VEVENT
BEGIN:VEVENT
SUMMARY:OpenCL Ecosystem Updates
DTSTART;VALUE=DATE-TIME:20220620T080000Z
DTEND;VALUE=DATE-TIME:20220620T084000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3181@indico.wigner.hu
DESCRIPTION:Speakers: Máté Nagy-Egri (Stream HPC)\nOpenCL has gained a n
ame for being one of the most portable programming models for heterogenous
programming. It’s 3.0 release improves upon deployment flexibility and
most major vendors already ship 3.0 compliant runtimes. Stream HPC is at t
he forefront of reinvigorating the ecosystem\, demonstrating how to make t
he best use of 3.0 and the latest features. The OpenCL SDK is a 1st party
Khronos SDK aimed at being the „one-stop-shop” for OpenCL developers.
A feature that was released close to 3.0 is the ability to register OpenCL
Layers between applications and OpenCL drivers. Stream HPC also took part
in implementing validation layers for OpenCL to help developers catch err
onous code with informative diagnostics before/instead of crashing your ap
plication.\n\nhttps://indico.wigner.hu/event/1393/contributions/3181/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3181/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Custom Tailored FPGA Boson Sampling
DTSTART;VALUE=DATE-TIME:20220621T075000Z
DTEND;VALUE=DATE-TIME:20220621T081000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3166@indico.wigner.hu
DESCRIPTION:Speakers: Gregory Morse (Department of Programming Languages a
nd Compilers\, Eötvös Loránd University)\nComputing the permanent of a
matrix finds an important application in the context of boson sampling. U
sing the BB/FG permanent formula with a reflected binary Gray code\, we im
plemented an FPGA design aimed at maximizing the use of logic and DSP reso
urces to increase the parallelism and reducing the time complexity from $\
\mathcal{O}(n.2^{n-1})$ to $\\mathcal{O}(n.2^{n-3})$. This can be further
ed reduced by half using both cards of the FPGA in a dual array mode of op
eration. To allow the design to scale up to 40x40 matrices and achieve a
speed of 280MHz\, we used a properly pipelined state-of-the-art rectangula
r multiplier algorithm to match the structure of the DSP units of the FPGA
.\n\nIn practical boson sampling configurations\, there will be photons sh
aring an optical mode\, which mathematically refers to computing the perma
nent of a matrix with independent repeated rows and repeated columns. Thi
s allows for a generalization of BB/FG using binomial coefficients to simp
lify multiplicities across rows or columns:\n\\begin{equation}\n \\text
{rp}(\\boldsymbol{A}\, \\boldsymbol{M}\, \\boldsymbol{N})=\\frac{1}{2^{n-1
}} \\sum\\limits_{\\boldsymbol{\\Delta}} \\bigg(\\prod\\limits_{k=1}^{m}
(-1)^{\\Delta_k} {M_k \\choose \\Delta_k}\\bigg) \\prod\\limits_{j=1}^{m}
\\bigg(\\sum\\limits_{k=1}^{m}\n \\left(M_k-2\\Delta_k\\right)a_{k\
, j}\\bigg)^{N_j}\n\\end{equation}\n where $A$ is a square matrix describi
ng the interferometer with $m$ modes\, $\\boldsymbol{M}$ and $\\boldsymbol
{N}$ are the row and column multiplicities respectively such that the phot
on count $n=\\sum\\limits_{r\\in R}=\\sum\\limits_{c\\in C}$ and $\\Delta$
is an n-ary Gray code\, required for efficient computation. Computing bi
nomial coefficients efficiently presents design challenges on the FPGA. B
y constructing a large enough loop\, this can be resolved but it requires
special logic around the n-ary Gray code. Certain techniques like Guan co
des are insufficient as they lack the reflection property being computable
with simple logic so we based our method on a dynamic programming techniq
ue. We extended this approach to stagger the Gray code at precise even in
tervals based on the loop length\, incurring a constant initialization sum
mation cost\, and then a streamlined operation. To avoid division\, we im
plemented division via multiplication by "magic numbers". To account for
the BB/FG "anchor" row and allow the $4$ parallel operations to proceed wi
th a simultaneous "smooth" same row update\, we reduce the smallest multip
licity up to three times to maintain optimal complexity.\n\nAll our implem
entations were designed to automatically reset on completion\, providing f
or batching capability\, an important optimization for FPGA designs and fi
tting perfectly into the context of boson sampling\, where many permanents
are computed and the batch size equals the matrix size. As we used fixed
point arithmetic\, we conducted accuracy testing against a CPU infinite p
recision calculator. We benchmarked against similar maximally efficient i
mplementations on CPU. These are a part of the piquassoboost extension to
the piquasso library. We measured two important metrics against the CPU:
the matrix size cross-over threshold due to FPGA initialization time dela
y and performance speed-up. For non-repeated permanents\, we achieve an 8
.86x speed-up over CPU with a cross-over threshold at 16x16 matrices\, whi
ch batching reduces it to 9x9 matrices. For repeated row/column permanents
\, it is dependent upon the number of photons. For 20 photons\, the speed
up is 5.9x and cross-over 25x25\, while with batching the speed-up is 15x
and the cross-over is 15x15. Our effective equivalent if long double flo
ating point operations were used is $\\frac{(C_A+C_M)*280*10^6}{10^9}$ whe
re $C_A=2A=2*(40+4)$ and $C_M=4M+2A=6*4*39$ represent complex addition and
multiplication respectively\, yielding $285.5$ GFLOPS.\n\nhttps://indico.
wigner.hu/event/1393/contributions/3166/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3166/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Simulating gold resonant nano-antennas for nano-fusion
DTSTART;VALUE=DATE-TIME:20220621T120000Z
DTEND;VALUE=DATE-TIME:20220621T123000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3178@indico.wigner.hu
DESCRIPTION:Speakers: Istvan Papp (Wigner FK)\nModern theoretical fusion r
esearch is powered by plasma simulations\, inertial confinement fusion usu
ally involves particle-in-cell (PIC) methods containing lasers interacting
with charged particles. Here we show a simple kinetic model of resonant g
old nanoantennas both in vacuum and a monomer used as fillings in dentistr
y. The model manages to describe well qualitatively and quantitatively the
behaviours of nanoantennas while taking into account additional existing
effects\, such as electron spillout and eventually the destruction of the
nanorods. We will study the nanoantenna's lifetime and absorption properti
es.\n\nhttps://indico.wigner.hu/event/1393/contributions/3178/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3178/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Parallel computing for determining stable parameter domain in mech
atronic applications
DTSTART;VALUE=DATE-TIME:20220621T125000Z
DTEND;VALUE=DATE-TIME:20220621T131000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3155@indico.wigner.hu
DESCRIPTION:Speakers: Tamás Haba (Budapest University of Technology and E
conomics)\nThe field of mechatronics engineering integrates mechanical sys
tems and control\; therefore\, the main challenges of both subfields appea
r simultaneously. A typical task in mechatronics is position control\, whe
re the main goal is to reach the desired position or track a predefined tr
ajectory. The primary design task is to determine the stability domain of
the control parameters where the emerging vibrations converge to a stable
equilibrium. A mechatronic system consists of a mechanical structure and d
igitally implemented control\, resulting in combined effects of friction a
nd sampling. Both of them can cause non-smooth dynamics\, where the forces
have discontinuities. The friction force can have discontinuities at velo
city reversals\, while the control force changes at every sampling instant
. This dynamical behaviour makes it extremely hard to make the analysis an
alytically. Most of the time\, approximating models are used where the fri
ction effects are neglected. This approximation results in an inaccurate s
tability domain\, limiting the design process. The stability analysis can
be done with numerical simulation at each parameter combination\, taking t
he friction effects into account. Still\, this method is rarely used becau
se of its high computational costs when high-resolution stability charts a
re needed.\n\n\nIn this study\, a parallel computing approach is introduce
d for exploring the stability domain of mechatronics systems with simulati
ons. Parallelization and the general-purpose application of GPUs can radic
ally accelerate computing tasks where partial results can be computed simu
ltaneously. In this specific case\, simulations corresponding to different
parameter combinations can be run independently from each other\, allowin
g the possibility of parallelizing. GPU programs can have thousands of thr
eads\, making the simulations extremely efficient without the accuracy tra
de-off of different methods.\n\n\nThe study presents a GPU based method fo
r determining the stability domain of a mechatronic system through an exam
ple of position control. The basic model of the system with Coulomb fricti
on and the control law is shown\, and a discrete-time model is presented.
The simulations based on the mapping are implemented in OpenCL and tested
on GPU. Results show that the proposed method efficiently produces high-re
solution stability charts.\n\nhttps://indico.wigner.hu/event/1393/contribu
tions/3155/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3155/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Exploring SARS-CoV-2 receptor binding domain variants
DTSTART;VALUE=DATE-TIME:20220621T100000Z
DTEND;VALUE=DATE-TIME:20220621T103000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3161@indico.wigner.hu
DESCRIPTION:Speakers: Ákos Gellért ()\nThe COVID-19 epidemic created an
extraordinary situation for the whole humanity\, claiming millions of live
s and causing a significant economic setback. At the same time\, the inter
national research community has rapidly generated an order of magnitude la
rger data set than ever before\, which can contribute to understanding the
evolution and dynamics of the epidemic\, to its containment and to the pr
evention of similar pandemics. Although genetic sequences are available in
a never before seen amount (as of April 2022\, more than 10 million compl
ete sequences at GISAID) a key question is what kind of phenotypic changes
the mutations cause and if we can estimate the virulence or severity of a
certain variant solely from the sequence. Recently a so-called "deep muta
tional scanning" database became available in which receptor binding affin
ities are measured for tens of thousands of mutations in several variants.
We will calculate the 3D structure to these variants with GPU accelerated
AlphaFold software and prepare the resulting database for further machine
learning analysis.\n\nhttps://indico.wigner.hu/event/1393/contributions/3
161/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3161/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Polynomial speedup in exact Torontonian calculation by a scalable
recursive algorithm
DTSTART;VALUE=DATE-TIME:20220621T092000Z
DTEND;VALUE=DATE-TIME:20220621T094000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3154@indico.wigner.hu
DESCRIPTION:Speakers: Ágoston Kaposi ()\nEvaluating the Torontonian funct
ion is a central computational challenge in the simulation of Gaussian Bos
on Sampling (GBS) with threshold detection.\nDuring the calculation of thi
s matrix function exponentially large number of determinants have to be co
mputed.\nWe proposed a recursive algorithm providing a polynomial speedup
in the exact calculation of the Torontonian compared to state-of-the-art a
lgorithms.\nOur algorithm recursively reuses the data used before to reach
the computational advantage.\nAccording to numerical analysis the complex
ity of the algorithm is ordo(n^1.06912 * 2^n).\nWith our algorithm\, one c
an simulate threshold GBS up to 35-40 photon clicks without the needs of l
arge-scale computational capacities.\n\nhttps://indico.wigner.hu/event/139
3/contributions/3154/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3154/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Improving efficiency of non-Gaussian photonic circuit simulations
DTSTART;VALUE=DATE-TIME:20220621T073000Z
DTEND;VALUE=DATE-TIME:20220621T075000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3156@indico.wigner.hu
DESCRIPTION:Speakers: Zoltán Kolarovszki ()\nThe simulation of photonic q
uantum computers with non-Gaussian circuit elements\nhas a high memory usa
ge since the quantum state is usually represented as a\ntensor\, which sca
les exponentially in the number of modes in the photonic\ncircuit. However
\, this representation turns out to be slow and overabundant in\nmost case
s\, forcing us to devise a new strategy for simulating general\nnon-Gaussi
an photonic circuits.\n\nIn our proposed strategy\, the way to cut off the
quantum state is more \neconomical in terms of data\, so that the memory
usage of the simulation can be \nsignificantly reduced. In our recently de
veloped simulator (called Piquasso) we \nimplemented this strategy\, which
enabled us to perform simulations faster than before.\n\nhttps://indico.w
igner.hu/event/1393/contributions/3156/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3156/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Efficient quantum gate decomposition via adaptive circuit compress
ion
DTSTART;VALUE=DATE-TIME:20220621T081000Z
DTEND;VALUE=DATE-TIME:20220621T084000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3152@indico.wigner.hu
DESCRIPTION:Speakers: Peter Rakyta (Department of Physics of Complex Syste
ms\, Eötvös Loránd University)\nIn this work\, we report on a novel qua
ntum gate approximation algorithm based on the application of parametric t
wo-qubit gates in the synthesis process. The utilization of these parametr
ic two-qubit gates in the circuit design allows us to transform the discre
te combinatorial problem of circuit synthesis into an optimization problem
over continuous variables. The circuit is then compressed by a sequential
removal of two-qubit gates from the design\, while the remaining building
blocks are continuously adapted to the reduced gate structure by iterated
learning cycles. We implemented the developed algorithm in the SQUANDER s
oftware package and benchmarked it against several state-of-the-art quantu
m gate synthesis tools. Our numerical experiments revealed outstanding cir
cuit compression capabilities of our compilation algorithm providing the m
ost optimal gate count in the majority of the addressed quantum circuits.\
n\nhttps://indico.wigner.hu/event/1393/contributions/3152/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3152/
END:VEVENT
BEGIN:VEVENT
SUMMARY:The highly increased number of protein structures calls for high p
erformance algorithms
DTSTART;VALUE=DATE-TIME:20220620T150000Z
DTEND;VALUE=DATE-TIME:20220620T152000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3165@indico.wigner.hu
DESCRIPTION:Speakers: Tamas Hegedus (Semmelweis University)\nThe number of
unique transmembrane (TM) protein structures doubled in the last four yea
rs that can be attributed to the revolution of cryo-electron microscopy. I
n addition\, the AlphaFold2 (AF2) deep learning algorithm also provided a
large number of predicted structures with high quality. However\, if a spe
cific protein family is the subject of a study\, collecting the structures
of the family members is highly challenging in spite of existing general
and protein domain-specific databases.\n\nWe demonstrate this and assess t
he applicability of automatic collection of protein structures via the ABC
protein superfamily. We developed a pipeline to identify and classify tra
nsmembrane ABC protein structures and also to determine their conformation
al states based on special geometric measures\, conftors. This and similar
processes need alignment of structures with a run time of 1-10s that was
feasible on the scale of experimental structures (n<300K). However\, the ~
100M theoretical\, high quality AF2 protein structures renders the calcula
tions challenging and requires reimplementation of various algorithms.\n\n
Since the AlphaFold database contains structure predictions only for singl
e chains\, we performed AF-Multimer predictions for human ABC half transpo
rters functioning as dimers. Our AF2 predictions warn of possibly ambiguou
s interpretation of some biochemical data regarding interaction partners a
nd call for further experiments and experimental structure determination.
In order to organize structural data and made novel structure predictions
and their annotation available for the broader scientific community\, we j
oined the 3D-Beacon Network community to develop data and API standards.\n
\nhttps://indico.wigner.hu/event/1393/contributions/3165/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3165/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Preliminary results of the tuned HIJING++ heavy-ion event generato
r
DTSTART;VALUE=DATE-TIME:20220620T144000Z
DTEND;VALUE=DATE-TIME:20220620T150000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3169@indico.wigner.hu
DESCRIPTION:Speakers: Balazs Majoros (Wigner FK)\nMonte Carlo event genera
tors became one of the most important tools of modern high-energy physics.
They are widely used in the high-energy community to simulate particle co
llisions\, make predictions\, and to design experiments.\nThe HIJING++ (He
avy Ion Jet INteraction Generator) is the successor of the 30 year old For
tran HIJING\, completely rewritten in C++\, providing multi-thread process
ing and various new modular features. In order to have meaningful data fro
m HIJING++ however\, it had to be tuned to reproduce existing experimenta
l data correctly. An important and resource-consuming phase of the develop
ment is the tuning of the internal parameters to reproduce the existing ex
perimental data. These parameters cannot be determined by direct methods o
f calculation\, therefore to get the desired\, optimal generator responses
we have to run the generator in every value combinations of these paramet
ers . This is the most computationally heavy part in the development of th
e generator. After several months of CPU time and hundreds of terabytes of
generated data we settled down on the internal parameters of the HIJING++
.\nIn this talk I want to highlight the process that was used to tune the
HIJING++ and show the results of this process. The current state of HIJING
++ can reliably reproduce experimental particle collision data\, in variou
s collision systems and energies.\n\nhttps://indico.wigner.hu/event/1393/c
ontributions/3169/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3169/
END:VEVENT
BEGIN:VEVENT
SUMMARY:ELKH Cloud vGPU challenges and implementation steps
DTSTART;VALUE=DATE-TIME:20220620T090000Z
DTEND;VALUE=DATE-TIME:20220620T093000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3173@indico.wigner.hu
DESCRIPTION:Speakers: Ádám Pintér ()\nhttps://indico.wigner.hu/event/13
93/contributions/3173/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3173/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Present and Future of GPU HPC in Hungary
DTSTART;VALUE=DATE-TIME:20220620T093000Z
DTEND;VALUE=DATE-TIME:20220620T095000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3163@indico.wigner.hu
DESCRIPTION:Speakers: Zoltan Kiss (KIFÜ)\nKomondor\, the newest 5 petaflo
ps GPU supercomputer of Hungary is on its way to both academic and industr
ial users. The system includes 200 A100 GPU. The talk will focus on the de
tails of the hardware and software stack\, and ways to access and use the
system\, including the option to run containerized jobs by using web inter
faces only.\nThe HPC Competence Center will offer training on how to use t
he newest infrastructure efficiently\, including GPU programming training
courses in English.\nKIFÜ is already planning the upgrade of the infrastr
ucture. The machine in question will offer 20 petaflops of capacity integr
ated into the EuroHPC ecosystem. KIFÜ intends to benefit from cutting-edg
e GPU technologies\, but would gladly hear feedback from the community\, t
o learn your opinions of future technologies and ways we would help you wi
th your tasks effectively. The system will be ready to be integrated with
Quantum computers.\n\nhttps://indico.wigner.hu/event/1393/contributions/31
63/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3163/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Parallel Implementation of Multivariate Empirical Mode Decompositi
on on GPU
DTSTART;VALUE=DATE-TIME:20220620T095000Z
DTEND;VALUE=DATE-TIME:20220620T101000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3158@indico.wigner.hu
DESCRIPTION:Speakers: Zeyu Wang (The University of Pannonia)\nEmpirical Mo
de Decomposition (EMD) is an effective tool for the analysis of non-linear
and non-stationary signals\, which has been widely used in various applic
ation fields for noise reduction\, feature extraction and classification.
Due to its adaptive and data-driven nature\, it has been introduced to ele
ctroencephalography (EEG) analysis to extract more accurate information in
time-frequency\, phase coherence and brain connectivity analysis. EMD met
hod decomposes signal into several narrow band oscillatory mode components
\, known as Intrinsic Mode Functions (IMFs). Despite its advantage and im
portance\, using EMD in signal processing is problematic as the algorithm
is computationally very expensive. For high-density\, high-resolution EEG
measurements\, the runtime can easily reach several hours.\n\nOver the pas
t decade\, several variants of the EMD method have been proposed\, includi
ng Multivariate Empirical Mode Decomposition (MEMD). MEMD focuses on the e
mpirical mode decomposition process of multi-channel signals\, it treats t
he input signal as a multivariate signal in a high-dimensional space. By p
rojecting the signal onto each direction vector and calculating the multiv
ariate envelopes and IMFs\, the synchronous decomposition of the multichan
nel signal can be realized. However\, multi-channel signals will bring a h
eavier workload which makes MEMD computationally even more expensive. \n\n
In this talk\, we will describe the implementation strategy and details of
a parallel CUDA MEMD algorithm. We will start with an overview of the num
erical steps of MEMD\, then the details of the parallelization steps\, inc
luding direction vector generation\, signal projection\, extrema detection
and selection\, and the cubic spline interpolation\, etc. Compared with M
EMD implementation in the MATLAB-based EMDLAB toolbox\, our GPU parallel v
ersion achieves about 150x performance improvement reducing execution time
from hours to minutes.\n\nhttps://indico.wigner.hu/event/1393/contributio
ns/3158/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3158/
END:VEVENT
BEGIN:VEVENT
SUMMARY:Accelerating massively parallel .NET code using FPGAs with Hastlay
er
DTSTART;VALUE=DATE-TIME:20220620T072000Z
DTEND;VALUE=DATE-TIME:20220620T080000Z
DTSTAMP;VALUE=DATE-TIME:20230204T111457Z
UID:indico-contribution-1393-3168@indico.wigner.hu
DESCRIPTION:Speakers: Zoltán Lehóczky (Lombiq Technologies Ltd.)\nHastla
yer (https://hastlayer.com/) by Lombiq Technologies is a .NET software dev
eloper-focused\, easy-to-use high-level synthesis tool with the aim of acc
elerating massively parallel applications. It converts standard .NET Commo
n Intermediate Language (CIL) bytecode into equivalent Very High Speed Int
egrated Circuit Hardware Description Language (VHDL) constructs which can
be implemented in hardware using FPGAs. Cloud-available FPGA platforms are
supported for high-performance use-cases\, as well as the Zynq 7000 famil
y of FPGA SoC devices for low-power embedded systems like drones and nanos
atellites. In this talk\, we'll introduce Hastlayer and how it can be used
\, our results showing up to 2 orders of magnitude speed and power efficie
ncy increases\, as well as the collaboration partners we seek from academi
a and other industry players.\n\nhttps://indico.wigner.hu/event/1393/contr
ibutions/3168/
LOCATION:Hotel Mercure Budapest Castle Hill
URL:https://indico.wigner.hu/event/1393/contributions/3168/
END:VEVENT
END:VCALENDAR