У оквиру семинара Центра за изучавање комплексних система Института за физику у Београду, у четвртак, 18. фебруара 2021. године у 14 часова путем Zoom платформе, др Милан Радоњић (Лабораторија за примену рачунара у науци, Институт за физику у Београду, Департман за физику, Технички универзитет Кајзерслаутерн, Немачка) одржаће предавање:

Non-equilibrium evolution of Bose-Einstein condensate deformation in temporally controlled weak disorder

САЖЕТАК:

In this talk we will consider a time-dependent extension of a perturbative mean-field approach to the homogeneous dirty boson problem by investigating how switching on and off a weak disorder potential affects the stationary state of an initially equilibrated Bose-Einstein condensate by the emergence of a disorder-induced condensate deformation. In the switch-on scenario, we will study if the stationary condensate deformation turns out to be a sum of an equilibrium part [1], that actually corresponds to adiabatically switching on the disorder, and a dynamically-induced part, where the latter depends on the particular driving protocol [2]. If the disorder is switched off afterwards, we will find out if the resulting condensate deformation acquires an additional dynamically-induced part in the long-time limit, and what happens with the equilibrium part. We will also present an appropriate generalization to inhomogeneous trapped condensates. Our results demonstrate that the condensate deformation represents an indicator of the generically non-equilibrium nature of steady states of a Bose gas in a temporally controlled weak disorder.

[1] K. Huang and H.-F. Meng, Phys. Rev. Lett. 69, 644 (1992)
[2] Milan Radonjić and Axel Pelster, SciPost Phys. 10, 008 (2021)

Приступите предавању

Meeting ID: 839 8287 2129
Passcode: 529607

У оквиру семинара Групе за гравитацију честице и поља Института за физику у Београду, у петак, 12. фебруара 2021. године у 11 часова путем BigBlueButton платформе, др Игор Салом (Група за гравитацију, честице и поља, Институт за физику у Београду) одржаће предавање:

Ортосимплектичка супералгебра као просторно-временска симетрија

САЖЕТАК:

Објаснићемо везу ортосимплектичке алгебре осп(1|8) са Поинкаре и конформном (супер)симетријом и показати зашто се ова симетрија још зове и генерализована суперконформна симетрија. Продискутоваћемо унитарне иредуцибилне репрезентације ове алгебре, а затим се сконцентрисати на најједноставније од ових репрезентација. Видећемо да најједноставнија репрезентација одговара простору безмасене релативистичке честице која, у зависности од хелицитета, аутоматски и неизбежно задовољава одговарајуће једначине кретања (нпр. Клајн-Гордонову, Диракову, односно Максвелове једначине). Посебно ћемо обратити пажњу на то како се симетрија е.м. дуалности појављује у овом контексту. Укратко ћемо размотрити и (по комплексности) прву следећу репрезентацију и видети да она одговара честицама са масом, при чему се јављају два масена члана, међусобно повезана симетријом е.м. дуалности.

Приступите предавању

У оквиру семинара Центра за изучавање комплексних система Института за физику у Београду, у четвртак, 11. фебруара 2021. године у 14 часова путем Zoom платформе, др Владимир Лончар (Лабораторија за примену рачунара у науци, Институт за физику у Београду) одржаће предавање:

hls4ml: Fast inference of deep neural networks in FPGAs

САЖЕТАК:

With edge computing, real-time inference of deep neural networks (DNNs) on custom hardware has become increasingly relevant. Smartphone companies are incorporating Artificial Intelligence (AI) chips in their design for on-device inference to improve user experience and tighten data security, and the autonomous vehicle industry is turning to application-specific integrated circuits (ASICs) to keep the latency low. While the typical acceptable latency for real-time inference in applications like those above is O(1) ms, other applications require sub-microsecond inference. For instance, high-frequency trading machine learning (ML) algorithms are running on field-programmable gate arrays (FPGAs), highly accurate devices, to make decisions within nanoseconds. At the extreme inference spectrum end of both the low-latency (as in high-frequency trading) and limited-area (as in smartphone applications) is the processing of data from proton-proton collisions at the Large Hadron Collider (LHC) at CERN. Here, latencies of O(1) microsecond are required and resources are strictly limited. To address these challenges, we have developed hls4ml, an open-source library that converts pre-trained ML models into FPGA firmware, targeting extreme low-latency inference in order to stay within the strict constraints imposed by the CERN particle detectors.

In this talk, we will describe the essential features of the hls4ml workflow and network optimization techniques, including how to reduce the footprint of a machine learning model using state-of-the art techniques such as model pruning and quantization through quantization aware training.

Приступите предавању

Meeting ID: 857 9507 6891
Passcode: 738237

У оквиру семинара Групе за гравитацију, честице и поља Института за физику у Београду, у петак, 29. јануара 2021. године у 11 часова путем BigBlueButton платформе, Тијана Раденковић (Група за гравитацију, честице и поља, Институт за физику у Београду) одржаће предавање:

Gauge symmetry of the 3BF theory for a generic Lie 3-group

САЖЕТАК:

The higher category theory can be employed to generalize the
BF action to the so-called 3BF action, by passing from the notion of a
gauge group to the notion of a gauge 3-group. In this work we determine
the full gauge symmetry of the 3BF action. To that end, the complete
Hamiltonian analysis of the 3BF action for a general Lie 3-group is
performed, by using the Dirac procedure. This analysis is the first step
towards a canonical quantization of a 3BF theory. This is an important
stepping-stone for the quantization of the complete Standard Model of
elementary particles coupled to Einstein-Cartan gravity, formulated as a
3BF action with suitable simplicity constraints. We show that the
resulting gauge symmetry group consists of the already familiar G-, H-,
and L-gauge transformations, as well as additional M- and N-gauge
transformations, which have not been discussed in the existing
literature.

ArXiv ePrint: 2101.04049

Приступите предавању

У оквиру семинара Центра за изучавање комплексних система Института за физику у Београду, у четвртак, 21. јануара 2021. године у 14 часова путем Zoom платформе, Ана Вранић (Лабораторија за примену рачунара у науци, Институт за физику у Београду) одржаће предавање:

Growth signals determine the topology of evolving networks

САЖЕТАК:

Network science provides us a theoretical framework for representing and studying various complex systems, including biological, technological, and social ones. These systems are composed of many units that interact with each other, and their collective behavior cannot be predicted from the behavior of individual elements. The structure of complex networks is essential for understanding the evolution and function of complex systems. Regardless of the different origins of complex networks in nature and society, it was shown that they share similar properties [1], such as broad degree distribution, degree-degree correlations, and they are clustered. Growing network models are often used for exploring the dynamics and topology of complex networks. Network growth, in combination with linking rules, shapes the topology of a network. For example, in the Barabási-Albert model [2], growth and preferential attachment lead to broad degree distribution networks. So far, the focus was mostly on various linking rules and their influence on network structure. The majority of the models assume that the network growth is constant, i.e., at each time step, one new node is introduced. However, the growth of real systems is anything but constant. It varies in time, has trends and cycles, and long-range temporal correlations [3].

In this talk, we will explore how time-varying growth influences the structure of evolving complex networks. We will consider the aging nodes [4] model and include time-varying network growth. We will use different real and computer-generated time-varying growth signals to generate complex networks. Afterwards, we will compare the structure of these networks with the ones obtained with constant growth signals, and show that the properties of the growth signal significantly determine the topology of the obtained networks. Our results indicate that time-varying growth should be considered as a parameter in models of complex systems [5].

[1] Boccaletti S., Latora V., Moreno Y., Chavez M., and Hwang D.U., Phys. Rep. 424, 175 (2006).
[2] Barabási A. L. and Albert R., Science 286, 509 (1999).
[3] Mitrović Dankulov M., Melnik R., and Tadić B., Sci. Rep. 5, 12197 (2015).
[4] Hajra K. B. and Sen P., Phys. Rev. E 70, 056103 (2004).
[5] Vranić A. and Mitrović Dankulov M., accepted in JSTAT, arXiv:2009.00444 (2020).

Приступите предавању

Meeting ID: 826 8858 6903
Passcode: 952042