У оквиру СЦЛ семинара Центра за изучавање комплексних система, у четвртак, 20. септембра 2018. у 14 часова у читаоници библиотеке „Др Драган Поповић“, др Андреја Стојић (Лабораторија за физику животне средине, Институт за физику у Београду) одржаће предавање:

Explainable machine learning

САЖЕТАК:

Over recent years, the success in the application of artificial intelligence implemented in machine learning, supported by the great availability of high-dimensional data, has become evident in all fields of science. Nevertheless, understanding and correctly interpreting complex models for predicting natural and social phenomena, such as random forests, deep neural networks or an extreme gradient boosting, can be challenging. The ability to accurately interpret a model’s prediction, comprehend causality and features that drive prediction, supports deeper understanding of the process being modeled. Therefore, it is critical that researchers gain insight into the way such models arrive at their predictions. This seminar will cover some basic concepts of machine learning and an a posteriori explanation framework based on consistent, locally accurate, individualized feature attribution methods, aimed at shedding light on problems where human intuition and domain knowledge are often limited.

У оквиру СЦЛ семинара Центра за изучавање комплексних система, у четвртак, 13. септембра 2018. у 14 часова у читаоници библиотеке „Др Драган Поповић“, проф. др Златко Папић (School of Physics and Astronomy, University of Leeds, UK) одржаће предавање:

Quantum many-body scars

САЖЕТАК:

Recent experiments on large chains of Rydberg atoms [H. Bernien et al., Nature 551, 579 (2017)] have demonstrated the possibility of realizing 1D systems with locally constrained Hilbert spaces, along with some surprising signatures of non-ergodic dynamics, such as persistent oscillations following a quench from the Neel product state. In this talk we will argue that this phenomenon is a manifestation of a „quantum many-body scar“, i.e., a concentration of extensively many eigenstates of the system around special many-body states [1]. The special states are analogs of unstable classical periodic orbits in the single-particle quantum scars. We will present a model based on a single particle hopping on the Hilbert space graph, which quantitatively captures the scarred wave functions up to large systems of 32 atoms. These results suggest that scarred many-body bands give rise to a new universality class of quantum dynamics, which opens up opportunities for creating and manipulating novel states with long-lived coherence in systems that are now amenable to experimental study.

[1] C. J. Turner, A. A. Michailidis, D. A. Abanin, M. Serbyn, Z. Papić, Nat. Phys, 14, 745 (2018).

У оквиру семинара Лабораторије за спектроскопију плазме и ласере, у понедељак, 10. септембра 2018. године у 13:30 часова у читаоници библиотеке „Др Драган Поповић“, Предраг Ранитовић ће одржати предавање:

Attosecond Coherent Control of Atomic and Molecular XUV and X-Ray
Absorption Processes by Intense Laser Radiation

САЖЕТАК:

Recent breakthroughs in the ultrafast lasers development have allowed for the opening of new horizons in the ultrafast experimental and theoretical AMO physics. Following these advances, table-top light sources have provided novel ways to achieve a real-time manipulation of the correlated electron/nuclear wavepacket dynamics by means of attosecond XUV and femtosecond IR radiation [1-6]. However, more than fifteen years upon the birth of attosecond physics, and after the Nobel Prize Award for femtochemistry, the concept of attosecond chemistry has not yet been fully realized. In this talk, I will introduce the basic principles of the strong laser-field ionization and generation of attosecond XUV/X-Ray radiation, and show how the attosecond VUV and XUV light sources can be used to coherently manipulate molecular dynamics by means of electron wave packet interferometry [7-9]. The use of attosecond VUV pulse-trains to coherently excite and control the outcome of a simple chemical reaction in a deuterium molecule, in a nonBorn-Oppenheimer regime, presents intriguing new possibilities for bridging the gap between the attosecond physics and attochemistry [10]. Furthermore, I will discuss very recent results that extend the use of the attosecond table-top techniques to the soft X-Ray regime in the water window [11].

[1] E. Gagnon, P. Ranitovic, X. M. Tong et al. Science 317(5843), 1374-1378 (2007).
[2] A. S. Sandhu, E. Gagnon, R. Santra, et al. Science 322 (5904), 1081-1085 (2008).
[3] P. Ranitovic, X. M. Tong, B. Gramkow et al. New Journal of Physics 12, 013008 (2010).
[4] D. Hickstain, P. Ranitoivc, S. Witte et al. Phys. Rev. Lett. 109, 073004 (2012).
[5] P. Ranitovic, X. M. Tong, C. W. Hogle et al. Phys. Rev. Lett. 106, 053002 (2011).
[6] X. Zhou, P. Ranitovic, C. W. Hogle et al. Nature Physics 8, 232-237 (2012).
[7] P. Ranitovic, X. M. Tong, C. W Hogle et al. Phys. Rev. Lett. 106, 193008 (2011).
[8] C. Hogle, X. M. Tong, L. Martin, M. M. Murnane, H. C. Kapteyn and P. Ranitovic. Phys. Rev. Letters 115 (17), 173004 (2015).
[9] P. Ranitovic et al. Phys. Rev A, 98 (1), 013410 (2018).
[10] P. Ranitovic et al. Proceedings of National Academy of Sciences, 111(3), pp. 912-917 (2014).
[11] Yoann Pertot, et al. Science 355, 6322 (2017).

Short Biography: Predrag Ranitovic got his PhD from University of Stockholm. His thesis was based on the experimental work performed at Kansas State University where he developed experiments for 3D momentum imaging of small atoms and molecules by attosecond XUV radiation. He continued developing methods for attosecond coherent control of electron and molecular dynamics in the Kapteyn-Murnane group
at University of Colorado and JILA at Boulder. At JILA, he also performed electron holography experiments by utilizing strong mid-IR laser radiation. After UC-Boulder, he moved to Lawrence Berkeley National Lab where he was applying time-resolved femtosecond IR and attosecond XUV methods to coherently control electron dynamics in nano-materials. During his stay at LBNL he was a group leader at ELI-ALPS – the first
large-scale laser user facility currently built in EU. Presently, Predrag is on sabbatical as a senior scientist at ETH – Zurich, and an LBNL affiliate.

У оквиру семинара Центра за чврсто стање и нове материјале, у петак, 24. августа 2018. године у 14.00 часова у читаоници библиотеке „Др Драган Поповић“ Института за физику у Београду, др Александар Матковић (Institute of Physics, Montanuniversität Leoben, Austria) ће одржати предавање:

„Charge transfer, charge trapping, and light-assisted charge spreading at the interfaces between organic semiconductors and 2D materials“

У оквиру традиционалног колоквијума Института за физику, у среду, 29. августа 2018. године у 12.00 часова у сали „Звонко Марић“, проф. др Тина Елиаси-Рад (College of Computer and Information Science, Northeastern University, Boston, USA) ће одржати предавање:

Graph distance from the topological view of non-backtracking cycles

САЖЕТАК:

Whether comparing networks to each other or to random expectation, measuring dissimilarity is essential to understanding the complex phenomenon under study. However, determining the structural dissimilarity between networks is an ill-defined problem, as there is no canonical way to compare two networks. Indeed, many of the existing approaches for network comparison differ in their heuristics, efficiency, interpretability, and theoretical soundness. Thus, having a notion of distance that is built on theoretically robust first principles and that is interpretable with respect to features ubiquitous in complex networks would allow for a meaningful comparison between different networks. In this talk, we will introduce a theoretically sound and efficient new measure of graph distance, based on the length spectrum function from algebraic topology, which compares the structure of two undirected, unweighted graphs by considering their non-backtracking cycles. We will show how this distance relates to structural features such as presence of hubs and triangles through the behavior of the eigenvalues of the non-backtracking matrix, and showcase its ability to discriminate between networks in both real and synthetic data sets. By taking a topological interpretation of non-backtracking cycles, this work presents a novel application of topological data analysis to the study of complex networks.