Spring 2014

23/01/ 2014

Recent Trends and Future Challenges in Action Recognition

Guest Speaker: Ivan Laptev is a research director at INRIA Paris-Rocquencourt, France. He received his PhD degree in Computer Science from the Royal Institute of Technology (KTH) in 2004 and a Master of Science degree from the same institute in 1997. He was a research assistant at the Technical University of Munich (TUM) during 1998-1999. He has joined INRIA as a postdoc in 2004 and became a full-time INRIA researcher in 2005. Ivan’s main research interests include visual recognition of human actions, objects and interactions. He has published over 50 papers at international conferences and journals of computer vision and machine learning. He serves as an associate editor of International Journal of Computer Vision and Image and Vision Computing Journal, he was/is an area chair for CVPR 2010, ICCV 2011, ECCV 2012, CVPR 2013 and ECCV 2014, he has co-organized several workshops and tutorials on human action recognition at major computer vision conferences. He has also co-organized a series of INRIA summer schools on computer vision and machine learning (2010-2013). Ivan was awarded ERC Starting Grant in 2012

Abstract: This computer vision talk will overview recent progress and open challenges in human action recognition. Specifically, I will focus on the three problems of (i) action representation in video, (ii) weakly-supervised action learning and (iii) ambiguity of action vocabulary. To the first problem, I will overview local feature methods providing state-of-the-art results on current action recognition benchmarks. Motivated by the difficulty of large-scale video annotation, I will next present our recent work on weakly-supervised action learning from video and corresponding video scripts. I will finish by highlighting limitations of the standard action classification paradigm and will show recent work addressing this problem

Video: http://www.youtube.com/watch?v=ht8K6K_TGjM


Numerical Algorithms for High-Dimensional Problems

Speaker: Ivan Oseledets is an Associate Professor at the Skolkovo Institute of Science of Technology (Skoltech) from 2013. He previously was a Senior Researcher in the Institute of Numerical Mathematics of Russian Academy of Sciences (INM RAS) in Moscow, Russia (till 2013, from 2013-part time). Ivan has received the PhD degree in Numerical Mathematics from the INM RAS in 2007 and the degree of Doctor of Sciences (second Russian degree) in 2012. Ivan was an invited Professor in the Haussdorf Institute of Mathematics in Bonn in 2011, and also has a part-time position in the Max-Planck Institute for Mathematics in the Sciences (Leipzig) from 2009. Dr. Oseledets research interests include numerical analysis, linear algebra, tensor methods and their applications in high-dimensional problems: solution of PDEs, quantum chemistry and computational material design, stochastic partial differential equations, wavelets, data mining and compression. Dr. Oseledets has received the medal of Russian academy of Sciences for the best student work in Mathematics in 2005; the medal of Russian academy of Sciences for the best work among young mathematicians in 2009. He is the winner of the Dynasty Foundation contest among young mathematicians in Russia in 2012.

Abstract: Multidimensional problems are notoriously difficult due to the curse of dimensionality. However, high-dimensional problems are usually the most interesting ones and moreover, if the problem is of a considerable practical interest, often there is a method that solves it. The most vivid example is the Schrodinger equation in quantum chemistry, where efficient solution methods have been proposed. However, such methods are usually problem-specific, require a lot of efforts to implement and difficult to be applied in other areas. In the recent years, active development of mathematical foundations for the algorithms for the solution of high-dimensional problem has begun. Novel tensor formats (Hierarchical Tucker, Tensor Train) as well as surprinsing connections with other research areas (MPS, PEPS, tensor networks, graphical models) form a new research area with new fascinating theoretical and algorithmic problems and new applications in chemistry, biology and data-mining and global optimization.

Where: Institute of Gene Biology RAS, Vavilova street 34/5, Conference room, 1st floor

Video: http://www.youtube.com/watch?v=9qeOKB_1Mh0


Fluctuations and Mixing in Active Suspensions

Guest speaker: Dr Dmitri (Mitya) O Pushkin graduated from the Moscow Institute of Physics and Technology in 1996. After a yearlong internship at the ABB Research Centre, Switzerland he joined a PhD program at the Department of Theoretical and Applied Mechanics of the University of Illinois at Urbana-Champaign (TAM, UIUC). His research focused on fundamental problems of cluster aggregation and on complex fluids. In particular, for a broad class of scale-free cluster aggregation processes that includes coagulation of colloidal particles, rain formation, growing networks and even bank mergers, he derived the steady self-similarity spectra. His PhD thesis won the best thesis award from TAM, UIUIC in 2004. Upon graduation Dmitri returned to his native Minsk, Belarus and had eventually become a business development director for an engineering IT company, Applied Systems Ltd. However, in 2010 he resumed active research at the Microgravity Research Centre, Université Libre de Bruxelles and moved to The Rudolf Peierls Centre for Theoretical Physics, University of Oxford, a year later. His recent results that include a novel physical mechanism of self-organisation of small particles in time-periodic flows and elucidating fundamentals of mixing in suspensions of microogranisms have been published in prestigious journals, such as Physical Review Letters, and led him to being an invited speaker at a number of international conferences in Europe and the United Kingdom

Abstract: Suspensions of active particles, such as swimming microorganisms, turn out to be efficient stirrers of the surrounding fluid. This fact may be directly relevant to the feeding and evolutionary strategies of swimming cells. Microfluidic devices exploring swimmers-induced mixing have been proposed. The possibility of a significant biogenic contribution to the ocean circulation is currently under intense debate. However, understanding fluctuations and the effective tracer diffusion coefficient in such non-equilibrium systems remains a challenge for modern theoretical physics. In this talk we focus on the fundamentals of these processes. We start by explaing why the theoretical approach based on the notion of the effective temperature of the bacterial bath and the ideas of the fluctuation–dissipation theorem, such as the Stokes–Einstein relation, breaks down in active suspensions. Next, we discuss the impediments to stirring by force-free swimmers and give a classification of possible stirring mechanisms. We show that fluid entrainment by individual swimmers and the effects of their trajectories curvature give rise to independent mixing mechanisms. We discuss their relative strength in dilute suspensions of active swimers and derive a simple expression for the effective tracer diffusion coefficient as a function of the swimmer parameters. During the discussion we demonstrate an interplay of topological, fluid mechanical and statistical physics concepts, such as closedness of tracer trajectories, the net fluid volume displaced by a moving body (also known as the Darwin drift) and properties of non-Gaussian random walks.

Where: Institute of Gene Biology RAS, Vavilova street 34/5, Conference room, 1st floor

Video: http://www.youtube.com/watch?v=GtfRaCHgeQ0


In vivo RNAi and Functional Genomics

Speaker: Professor Victor Kotelianski, Skoltech CREI Director on RNA Therapeutics.

Victor received his Doctor of Science (D.Sc.) degree from the National Cardiology Research Center, Academy of Medical Science in Moscow, and then went on to serve as the Institute’s Professor of Cell Biology. He was senior scientist at the Institute of Protein Research, National Academy of Science in Poustchino (Moscow Region). He received his Ph.D. in molecular biology from the National Academy of Sciences in Moscow and his M.D. from Uzhgorod University Medical School. He joined Alnylam in 2003 as Vice President of Research and was promoted to Senior Vice President, Distinguished Alnylam Fellow in 2008. He was formerly Distinguished Investigator, Director of Biological Research at Biogen (now Biogen Idec) and earlier in his career he served as Director of Research, Institut National de la Sante et de la Reserche Medicale (INSERM) in Paris. He has over 150 publications.

Abstract:  The talk will focus on the use of siRNAs in vivo to study gene function in the liver of mouse. In the first part of the talk, progress in siRNA delivery to mouse hepatocyte will be shown. The second part of the talk will be devoted to the analysis of three different cases of multiple mRNAs knockdown in studies of endocytosis, hypoxia and HYPPO pathway. RNA interference technology can now be used to examine the activities of specific genes in animal models.

Where: Institute of Gene Biology RAS, Vavilova street 34/5, Conference room, 1st floor

Video: http://youtu.be/1QmJc9TXfMQ


Fluid Modelling of Carbon Dioxide Sequestration

Guest speaker: Herbert Huppert, Professor of Theoretical Geophysics, University of Cambridge

Abstract:  Current global anthropogenic emissions of carbon dioxide are approximately 32 Gigatonnes annually.  The influence of this green-house gas on climate has raised concern.  A means of reducing environmental damage is to store carbon dioxide somewhere until well past the end of the fossil fuel era.  Storage by injection of liquid, or supercritical, carbon dioxide into porous reservoir rocks, such as depleted oil and gas fields and regional saline aquifers, is being considered.  The presentation will discuss the rate and form of propagation to be expected and quantify some of the risks involved. The talk builds on theoretical and experimental investigations of input of liquid of one viscosity and density from a point source above an impermeable boundary, either horizontal or slanted, into a heterogeneous porous medium saturated with liquid of different viscosity and density.  In the Sleipner natural gas field, carbon dioxide has been injected at a rate of ~ 1 Mt/yr since 1996.  We will briefly show how to apply our results to interpret these field observations.  One of the best controlled field experiments, the Otway Project, commenced on 2 April 2008 in Victoria, Australia.  Approximately sixty thousand tonnes of carbon dioxide was injected into a slanted sill over a period of just over a year.  We will show how accurately some of our theoretical models predict the field data obtained so far.  The talk will be illustrated by colour movie sequences of laboratory experiments and some simple desk-top demonstrations of aspects of flow of multi-phase fluids into a porous ambient.

Where: Institute of Gene Biology RAS, Vavilova street 34/5, Conference room, 1st floor

Video: http://www.youtube.com/watch?v=jfJ-CLRxzU


Encoding and Reasoning with Expert Knowledge in System Architecture: Applications in Satellite System

Guest speaker: Daniel Selva, Assistant Professor at Cornell University. Dr. Daniel Selva received a PhD in Space Systems from MIT in 2012, and he is currently finishing his post-doctoral appointment in the MIT System Architecture Lab and starting as an Assistant Professor at the Sibley School of Mechanical and Aerospace Engineering at Cornell University. His research interests focus on the application of knowledge engineering, global optimization and machine learning techniques to space systems engineering and architecture. Prior to MIT, Daniel worked for four years in Kourou (French Guiana) as a member of the Ariane 5 Launch team. Daniel has a dual background in electrical engineering and aeronautical engineering, with degrees from Universitat Politecnica de Catalunya in Barcelona, Spain, and Supaero in Toulouse, France. He is a 2007 la Caixa fellow, and received the Nortel Networks prize for academic excellence in 2002.

Abstract: Every system has an architecture: its essence, a high-level abstraction of its design. The process of choosing a system architecture is of critical importance: most of the performance and lifecycle cost of a system are fixed or committed by architectural decisions. And yet, most organizations still select system architectures based on unstructured, 100% human processes. This results in potentially sub-optimal architectures chosen from a very reduced set of candidate architectures, due to human limitations: bias, inconsistency in assessments, and low “computational speed”. On the other hand, much more structured, computer-aided processes are used in later phases of system design. These tools bring rigor, consistency, and exhaustiveness into the design process. Why can’t we use similar processes for system architecture? The answer is that system architecture is a much more open-ended, ill-posed problem that goes well beyond configuration design. This is the kind of task at which humans excel, and computers cannot do.

What is needed is a theory and a set of interactive tools that can combine the strengths of both humans and computers to find optimal system architectures. My research group at Cornell strives to address these problems by using concepts from logical reasoning systems, knowledge engineering, global optimization, and machine learning. I will discuss how the application of some of these techniques to two real-life complex system architecture problems in the domain of satellite systems for Earth observation, and communications brought forth interesting insights about these problems.

Where: Institute of Gene Biology RAS, Vavilova street 34/5, Conference room, 1st floor

Video: http://www.youtube.com/watch?v=dAqKJo__lyA


Chaos and Unpredictability  in Evolution

Guest speaker: Yaroslav Ispolatov,    Researcher – University of British Columbia

Abstract: Half a century ago, the discovery of deterministic chaos has revolutionized science. Surprisingly, few of these insights have entered the realm of evolutionary biology, where “survival of the fittest” epitomizes evolution as an optimization process that generally converges to equilibrium, the optimal phenotype. This perspective may be correct if only a simple phenotype, such as body size, had affected the selection, but in reality, all organisms have a multitude of phenotypic properties that impinge on their birth and death rates, and hence on evolutionary dynamics. However, evolution in high-dimensional phenotype spaces has rarely been considered.

Using the recently developed mathematical framework of adaptive dynamics we investigate the long-term evolutionary dynamics in simple single-spcecies competition models with high-dimensional phenotype spaces. Our main conclusion is that evolution is generally chaotic when
distinct phenotypic features affect the ecological interactions, such as competition for resources and predation, in complicated,  non-additive way.   These ecological interactions generate a evolutionary feedback loop, because selection pressure, which causes evolutionary change, changes itself as a population’s phenotype distribution evolves. Our conclusion supports Gould’s famous postulate, Wind back the tape of life to the early days of the Burgess Shale; let it play again from an identical starting point, and the chance becomes vanishingly small that anything like human intelligence would grace the replay”: if evolution is fundamentally chaotic, then evolution is generally unpredictable in the long term, even if the selection that drives it is deterministic. We also provide insights and estimates of how the probability of chaos in a generally dissipative dynamic systems
depends on its dimensionality.

Where: Institute of Gene Biology RAS,Vavilova street 34/5, Conference room, 1st  floor

Video: http://www.youtube.com/watch?v=gDkmv4Vhi2E


Many-Body Physics: Bose-Einstein Condensation and Superconductivity

Guest speaker: Professor Alexey Kavokin, Physics and Astronomy School, University of Southampton, Southampton (UK) and Russian Quantum Center, Moscow (Russia)

Abstract: The phenomenon of superconductivity has been discovered in 1911. Despite of over one hundred years of efforts and many Nobel prizes given to the research on superconductivity, still this remarkable effect can only be observed at low temperatures. All attempts to realize the room temperature superconductivity failed so far. Meanwhile, a similar phenomenon has been observed at a room temperature in 2007: the Bose-Einstein condensation of electrically neutral crystal quasiparticles, exciton-polaritons. At certain conditions tens of thousands of these quasiparticles may be accumulated in a single quantum state characterised by one fixed energy, phase and velocity. If the same would happen to charged particles, we would have a superconductivity at room temperature.In this talk I will analyse the possible ways of realisation of high temperature superconductivity mediated by Bose-Einstein condensates of excitons and exciton-polaritons. I will remind the concept of exciton mediated superconductivity proposed by Ginzburg and Bardeen in 1970s. The recent discoveries of high temperature Bose-Einstein condensation offer a new chance for realisation of this revolutionary concept.

Where: Institute of Gene Biology RAS, Vavilova street 34/5, Conference room, 1st floor

Video: http://www.youtube.com/watch?v=Qyy0hcE7Ihs


Mass Spectrometry of High Resolution for Biological Research

Speaker:  Professor Evgeny Nikolaev, Skoltech Founding Faculty Fellow, Head of Laboratory of Ion and Molecular Physics of the Institute for Energy Problems of Chemical Physics Russian Academy of Sciences, The Head of Laboratory of Mass Spectrometry of Biomacromolecules of the Institute for Biochemical Physics Russian Academy of Sciences, The main scientific researcher of the Institute Biomedical Chemistry Russian Academy of Medical Sciences

Abstract: Application of mass spectrometry in the field of top-down proteomics and protein modification analyses via accurate mass measurements is demanding further increase of the resolution and mass accuracy. We are discussing different approaches to MS analyses of biological mixtures and different instruments used: quadrupoles, time of flight, ion traps, ion cyclotron resonance MS (FT ICR) and orbitraps . Multidisciplinary project will be discussed which includes supercomputer simulations of ion cloud motion in high magnetic fields, ion cyclotron resonance spectrometry, signal processing, mass spectrometry in proteomics. Recently introduced new type of FT ICR measuring cell based on a Penning ion provides jump in both mass resolution and mass measurement accuracy. This new technology was born in supercomputer research of non neutral plasma behavior in Penning ion traps. Very high resolution was achieved for peptides (reserpine resolution 39 000 000 at 609 Da) and for proteins (BSA resolution 1 800 000 at 65 kDa) in moderate strength magnetic field (7 Tesla). Fine structure of isotopic cluster peaks in mass spectra of up to 5 KDa peptides has been resolved for the first time.

Video: http://www.youtube.com/watch?v=FuI_e5soBfY


Multiscale Modelling Based on Physical Analogies: from Aeroacoustics to Microfluidics

Guest speaker: Sergey Karabasov is a Senior Lecturer in Modelling and Simulation in School of Engineering and Materials Science Queen Mary University of London and a Royal Society University Research Fellow. Over the years, his research into the modelling of sound generated by aerodynamic flows has been supported by the leading UK industry such as Rolls-Royce, and also by the Engineering and Physical Sciences Research Council UK and the Royal Society of London. Internationally, his research into the modelling of sound generated by jet turbulence in collaboration with Ohio Aerospace Institute and NASA Glen is supported by Aero Acoustics Research Consortium (AARC). His other research interests include computational hydrodynamics for ocean modelling, multiscale methods for bridging computational fluid dynamics with atomistic simulations, and supercomputing. He is one of the developers of the high-resolution CABARET method. In 2013, CABARET and NAG were announced to be Winners of Sixth HPC Innovation Excellence Awards. Sergey holds a Full Doctorate of Science from Moscow Institute for Mathematical Modelling of the Russian Academy of Science. He is a Senior Member of American Institute of Astronautics and Aeronautics, Member of the Aeroacoustic Section of the Physical Acoustics Panel of the Russian Academy of Science, Member of the Royal Society Panel of Physical Sciences and a Guest Editor of the Royal Society PhilTrans A special issue “Multiscale systems in fluids and soft matter: approaches, numerics, and applications” which will be published in August 2014.

Abstract:  Multiscale problems, which solution requires multiscale methods, are encountered in many areas of applied science and engineering. One important ingredient of such methods is a technique for consistently coupling large scales with small scales in space and time that may differ by many orders of magnitudes within the same computational framework. The so-called acyclic techniques, where the information is exchanged one way only, e.g., from small scales to large scales, have been found efficient in a number of cases. An example of acyclic approach can be found in prediction schemes for noise generated by aerodynamic flows. In application to jet noise modelling for instance, the acyclic approach is used to link small-scale turbulence with large-scale acoustic waves in the framework of acoustic analogy. In other situations, where there is a considerable overlap between the small and the large scales, fully coupled (cyclic) approaches are required. For example, this is the case when the delivery of macromolecules (e.g. proteins) in microfluidic devices needs to be controlled by hydrodynamic shear and pressure gradient and when continuum hydrodynamics modelling and atomistic simulations must be performed concurrently. The cyclic approach we have developed for this kind of applications is based on a hydrodynamic analogy with two-phase flows. One ‘phase’ of the model is based on a continuum fluid dynamics model (Landau-Lifshitz Fluctuating Hydrodynamics equations) and the other one is based on pure atomistic representation (classical molecular dynamics). The ‘phases’ are coupled with preservation of macroscopic conservation laws. The interaction is governed by a numerical ‘zoom-in’ parameter to allow for a fully atomistic representation in the region of interest and a continuum representation where atomistic details are unnecessary. Examples of numerical implementation of the new coupling framework will be provided for the simulation of liquid argon (Leonnard-Jones potential) and water (Ben-Naim’s Mercedes Benz 2D water model).

Where: Institute of Gene Biology RAS, Vavilova street 34/5, Conference room, 1st floor

Video: http://www.youtube.com/watch?v=rd6VEnUqbJg&feature=youtu.be



High-Resolution Nonlinear Ion Mobility Separations with Mass Spectrometry and Selected Bioanalytical Applications

Guest Speaker: Dr. Alexandre Shvartsburg, Senior Scientist with the Biological Sciences Division,  Pacific Northwest National Laboratory (Richland, Washington, USA)

The main focus of Dr. Shvartsburg’s research is mass spectrometry (MS) and ion mobility spectrometry (IMS) coupled to MS, and applications of IMS/MS to the structural characterization of nanomaterials and biological macromolecules and rapid and selective isomer separations. Over the last few years, his interests have shifted to the novel IMS approach of field asymmetric waveform IMS (FAIMS) based on the difference between ion mobilities in strong and weak electric fields, and its bioanalytical applications. Advancing that technology has led him to conceptualize and explore novel nonlinear IMS methods that leverage the dependence of mobility on the field in new ways.

Abstract: Ion mobility spectrometry (IMS) is a method for analytical separation and characterization of ions, employing the properties of their transport through gases driven by electric field. The two branches of IMS are conventional or linear, based on the absolute ion mobility, and emerging differential or nonlinear, leveraging the dependence of mobility on the field strength. A major attraction of differential or field asymmetric waveform IMS (FAIMS) is high orthogonality to mass spectrometry (MS), far exceeding that of conventional IMS where the mobility is tightly correlated to the ion size and thus to mass. New gas compositions, elevated fields, and longer separations have raised the FAIMS resolving power from ~10 to ~100 – 500. This resolving power and orthogonality to MS have allowed FAIMS to distinguish many species, in particular structural isomers, co-eluting in conventional IMS. Examples include isomeric lipids with differing fatty acid attachment sites, peptides with sequence inversions or variant localization of post-translational modifications, and even isotopomers. Hydrogen-rich gas buffers have recently permitted separating protein conformers with extreme resolution, apparently isolating individual geometries for the first time by any solution or gas-phase technique. These unique capabilities open the door to previously unthinkable proteomic and metabolomic applications.

Where: Institute of Gene Biology RAS, Vavilova street 34/5, Conference room, 1st floor

Video: http://www.youtube.com/watch?v=h5t8tH7X19Y



Quantum Field Theory in Strong External Backgrounds

Guest Speaker:  Professor Dmitri M. Gitman, Head of Nuclear Physics Department, Institut of Physics,  University of São Paulo, Brazil

Abstract: QFT with an external background is an effective model when a part of a quantized field is strong enough to be treated as a given (external) classical one. Such a model is frequently used to describe quantum processed in the presence of different strong backgrounds like strong electromagnetic fields, gravitational fields, and so on. Especially interesting are backgrounds that violate the quantum vacuum stability, creating real particles from the vacuum. In our works a nonperturbative approach to such QFT’s models was elaborated. It is based on using exact solutions of relativistic wave equations in the corresponding backgrounds. QFT with an external background can be considered as a consistent model only if the back reaction of created particles is relatively small. We discuss this consistency on the example of two important cases of external fields in QED: a constant macroscopic electromagnetic field and the Coulomb field of a point like charge Ze.  To find the corresponding consistency restrictions on the macroscopic field and its duration, we calculate the change of mean energy density of quantized matter fields for an arbitrary constant electric field E, which is acting during a large but a finite time T.  In the case of the external Coulomb field, we dispute the common opinion that the Dirac equation with such a field with Z>137 it inconsistent. Defining rigorously self-adjoint extensions of Dirac operator with the Coulomb field for arbitrary Z, one can see that all they have nonsingular spectra and complete sets of orthonormalized wave functions. Recently, the graphene physics attracts a lot attention due to numerous applications in nanotechnology. It turns out the at certain conditions this physics can be effectively described by the so-called Dirac model, which is a QFT with essentially unstable vacuum under the action of the the external electric field. Applying our approach to study the electric conductivity in the graphene, we obtain results that match which existing experimentally observations and explain unusual behavior of this conductivity due to generation of charge carriers from the vacuum by the electric field.

Where: Institute of Gene Biology RAS, Vavilova street 34/5, Conference room, 1st floor

Video: http://www.youtube.com/watch?v=8EpStNhwcIw


Big-Data Induced Feedback Control Decision-Making in Competitive, Cooperative and Mixed Competitive-Cooperative Uncertain Environments

Guest speaker: Martin Jermakyan is an experienced researcher and financial engineer with both academic and applied credentials.  He has earned his PhD in Physics and Mathematics from the Mechanics and Mathematics department of Moscow State University.  Subsequently, he has been on the faculty of the Applied Mathematics Program at the Mathematics department of the University of California at Los Angeles and the Mathematics, Finance, and Operations and Industrial Research departments of the University of Michigan at Ann Arbor.  Having held senior level positions both at startups and large-corporations in the financial technological service industry in the USA, he brings more than 15 years of applied experience geared at the development and implementation of big-data processing applications in financial markets.

Abstract: Big-data processing analytics and its efficient implementation have increasingly become one of the demanded and most challenging topics of modern applied sciences and artificial intelligence in view of their applications, for example, in financial markets, electricity generation-transmission-distribution industry, healthcare, transportation logistics, multi-sensory systems, etc. At the same time, various big-data induced real-time mission critical applications in competitive, cooperative and mixed competitive-cooperative environments pose significant modeling, analytical, algorithmic, computational and logistic challenges as it pertains to the timely, reliable and optimal complex decision-making. The issues become even more acute in view of the 6V-attributes possessing big data – including those of veracity, validity and volatility of data. In competitively, cooperatively or mixed competitive-cooperatively interacting multiple big-data producing agent environments, where the agents of interaction recognize their environment from the observed data (by solving corresponding inverse problems) and learn further about the environment they operate in by producing additional observations through their actions (by utilizing variations of machine learning techniques), and where they adjust their actions to those of their competitors or cooperators, while their joint actions affect the state of the dynamic of the underlying system, most systematically optimal complex decisions can be modeled as Nash Feedback Equilibrium Control solutions of corresponding game-theoretical problems. Readily implementable theoretical results in this complex discipline are absent or scarce and, when available, the results merely represent case studies specifically prescribed with the purpose of arriving to relatively simple analytic solutions. However, hardly any applied big-data induced complex decision-making problems render themselves to such simplicity and susceptibility for solution. Furthermore, traditional methodologies frequently don’t leave sufficient time even for ad hoc decision-making. As a result and frequently, the only manageable solution can be achieved through the meta-algorithmisation of corresponding problems, accommodating generation of numerical solutions for all possible scenarios and with the organization of these solutions as layered tables which reside both in cache and various storage and database structures of specially designated decision-making sub-clouds. The talk will focus on big-data induced complex decision-making issues in competitive, cooperative or mixed competitive-cooperative environments and offer a mixture of known and new results.

Where: Hypercube, Skolkovo Innovation Center

Video: http://www.youtube.com/watch?v=QVKWyyJNF4o


Finite Time Singularities, Rogue Waves and Strong Collapse Turbulence

Guest speaker: Professor Pavel M. Lushnikov, Department of Mathematics and Statistics, University of New Mexico, Albuquerque, USA.

Pavel’s research interest includes a wide range of topics in applied mathematics, nonlinear waves and theoretical physics. Among them are laser fusion and laser-plasma interaction; dynamics of fluids with free surface, Kelvin-Helmholtz instability and nonlinear interactions of surface waves; theory of the wave collapse, singularity formation and its application to plasma physics, hydrodynamics, biology and nonlinear optics; bacterial aggregation, chemotaxis, cell-cell interactions. collapse of bacterial colonies, stochastic Potts model of biological cell; pattern formation in photorefractive crystals and other nonlinear optical media; high-bit-rate optical communication; dispersion-managed optical fiber systems; soliton propagation in optical systems; high performance parallel simulations of optical fiber systems; Bose-Einstein condensation of ultracold dipolar gases.

Abstract: Many nonlinear systems of partial differential equations have a striking phenomenon of spontaneous formation of singularities in a finite time (blow up). Blow up is often accompanied by a dramatic contraction of the spatial extent of solution, which is called by collapse. Near singularity point there is a qualitative change in underlying nonlinear phenomena, reduced models loose their applicability and other mechanisms become important such as inelastic collisions in the Bose-Einstein condensate, optical breakdown and dissipation in nonlinear optical media and plasma, wave breaking in hydrodynamics. Collapses occur in numerous reduced physical and biological systems including a nonlinear Schrodinger equation (NLSE) and a Keller-Segel equation (KSE).

We will focus on the collapse in the critical spatial dimension two (2D) which has numerous applications. For instance, 2D NLSE describes the propagation of the intense laser beam in nonlinear Kerr media (like usual glass) which results in the catastrophic self-focusing (collapse) eventually causing optical damage as was routinely observe in experiment since 1960-es. Recently such events have been also often referred as optical rogue waves. Another dramatic NLSE application is the formation of rogue waves in ocean. 2D KSE collapse describes the bacterial aggregation in Petri dish as well as the gravitational collapse of Brownian particles. We study the universal self-similar scaling near collapse, i.e. the spatial and temporal structures near blow up point. In the critical 2D case all these collapses share a strikingly common feature that the collapsing solutions have a form of either rescaled soliton (for NLSE) or rescaled stationary solution (for KSE). The time dependence of that scale determines the time-dependent collapse width L(t) and amplitude ~1/L(t).At leading order L(t)~ (t_c-t)^{1/2} for all mentioned equations, where t_c is the collapse time. Collapse however requires the modification of that scaling which in NLSE has the well-known loglog type ~ (\ln|\ln(t_c-t)|)^{-1/2} as well as KSE has another well-known type of logarithmic scaling modification. Loglog scaling for NLSE was first obtained asymptotically in 1980-es and later proven in 2006. However, it remained a puzzle that this scaling was never clearly observed in simulations or experiment. Similar situation existed for KSE. Here solved that puzzle by developing a perturbation theory beyond the leading order logarithmic corrections for both NLSE and KSE. We found that the classical loglog modification NLSE requires double-exponentially large amplitudes of the solution ~10^10^100, which is unrealistic to achieve in either physical experiments or numerical simulations. In contrast, we found that our new theory is valid starting from quite moderate (about 3 fold) increase of the solution amplitude compare with the initial conditions. We obtained similar results for KSE. In both cases new scalings are in excellent agreement with simulations. This efficiency of analytical results also allowed to study 2D NLSE-type dissipative system in the conditions of multiple random spontaneous formation of collapses in space and time. Dissipation ensures collapse regularization while collapses are responsible for non-Gaussian tails in the probability density function of amplitude fluctuations which makes turbulence strong. Power law of non-Gaussian tails is obtained for strong NLSE turbulence which is a characteristic feature of rogue waves. We suggest the spontaneous formation optical rogue from turbulent as a perspective route to the combing of multiple laser beams, generated by a number of fiber lasers, into a single coherent powerful laser beam.

Where: Institute of Gene Biology RAS, Vavilova street 34/5

Video: http://www.youtube.com/watch?v=hNAsT3Bu8Ws