We use cookies in order to improve the quality and usability of the HSE website. More information about the use of cookies is available here, and the regulations on processing personal data can be found here. By continuing to use the site, you hereby confirm that you have been informed of the use of cookies by the HSE website and agree with our rules for processing personal data. You may disable cookies in your browser settings.
25/12 Bol. Pecherskaya Ulitsa, Room 301
Nizhny Novgorod, 603155, Russia
Phone: (831) 416 95 40
email: svorontsova@hse.ru
Phone: (831) 416 98 00
email: ykuranova@hse.ru
In this paper, a user modeling task is examined by processing mobile device gallery of photos and videos. We propose a novel engine for preferences prediction based on scene recognition, object detection and facial analysis. At first, all faces in a gallery are clustered, and all private photos and videos with faces from large clusters are processed on the embedded system in offline mode. Other photos may be sent to the remote server to be analyzed by very deep sophisticated neural networks. The visual features of each photo are obtained from scene recognition and object detection models. These features are aggregated into a single descriptor in the neural attention unit. The proposed pipeline is implemented in mobile Android application. Experimental results for the Photo Event Collection, Web Image Dataset for Event Recognition and Amazon Fashion data demonstrate the possibility to efficiently process images without significant accuracy degradation.
The process of implementing agile technologies in an enterprise is an example of a highly demanding and challenging topic both for practitioners and academy. Speaking about agility, we basically mean various kinds of automation. A higher degree of automation results in a more agile enterprise. However, in practice, even in the case of complete automation of the enterprise, there remains a need for user interaction with various software systems. To increase the efficiency and simplify such interactions, it is necessary to develop mechanisms, which enable adaptation of user domain-specific interfaces for actual conditions of human-machine interaction. We consider for this purpose the use of an invariant formalism based on the allocation of stable, unchanging object data structures and interface structures. Evaluation of the effectiveness of the approach proposed is carried out with the example of the analytical system of the university admissions committee.
Cherenkov interaction between a wave pulse and a flow of electrons possessing a very wide (on the scale of the characteristic band of the resonant electron–wave interaction) velocity spread is considered. We show that if the wave pulse is short enough, and its group velocity is close to the phase velocity, then the effect of the slippage of the resonant electrons with respect to the wave pulse leads to the transformation of an inert electronic medium into an active one (absorbing or amplifying the wave pulse, depending on the slippage sign). This can be a mechanism of formation of short powerful electromagnetic pulses as a result of amplification of short-pulse weak noises by electron flows which, due to natural reasons, have a large velocity spread, namely, electron flows in the magnetosphere of planets, in the plasma envelope of brown dwarfs and neutron stars, as well as in electron masers with weak electron–wave interaction (including ultra-relativistic electron beams used in free-electron lasers).
In this study we consider the shortest path problem, where the arc costs are subject to distributional uncertainty. Basically, the decision-maker attempts to minimize her worst-case expected loss over an ambiguity set (or a family) of candidate distributions that are consistent with the decision-maker's initial information. The ambiguity set is formed by all distributions that satisfy prescribed linear first-order moment constraints with respect to subsets of arcs and individual probability constraints with respect to particular arcs. Under some additional assumptions the resulting distributionally robust shortest path problem (DRSPP) admits equivalent robust and mixed-integer programming (MIP) reformulations. The robust reformulation is shown to be NP-hard, whereas the problem without the first-order moment constraints is proved to be polynomially solvable. We perform numerical experiments to illustrate the advantages of the considered approach; we also demonstrate that the MIP reformulation of DRSPP can be solved effectively using off-the-shelf solvers.
The social dimension of IT-business inconsistency is one of the popular issues in the topic of IT-business misalignment. However, the studies that are used to develop a formalized approach use similar methodologies. Thus, the purpose of this article is to describe an approach based on the capabilities of conceptual analysis, focused on a formal search for misalignment
Two market network models are investigated. One of them is based on the classical Pearson correlation as the measure of association between stocks returns, whereas the second one is based on the sign similarity measure of association between stocks returns. We study the uncertainty of identification procedures for the following market network characteristics: distribution of weights of edges, vertex degree distribution in the market graph, cliques and independent sets in the market graph, and the vertex degree distribution of the maximum spanning tree. We define the true network characteristics, the losses from the error of its identification by observations, and the uncertainty of identification procedures as the expected value of losses. We use elliptically contoured distribution as a model of multivariate stocks returns distribution. It is shown that identification statistical procedures based on the sign similarity are statistically robust in contrast to the procedures based on the classical Pearson correlation
We suggest a universal map capable of recovering the behavior of a wide range of dynamical systems given by ODEs. The map is built as an artificial neural network whose weights encode a modeled system. We assume that ODEs are known and prepare training datasets using the equations directly without computing numerical time series. Parameter variations are taken into account in the course of training so that the network model captures bifurcation scenarios of the modeled system. The theoretical benefit from this approach is that the universal model admits applying common mathematical methods without needing to develop a unique theory for each particular dynamical equations. From the practical point of view the developed method can be considered as an alternative numerical method for solving dynamical ODEs suitable for running on contemporary neural network specific hardware. We consider the Lorenz system, the Rцssler system and also the Hindmarch – Rose model. For these three examples the network model is created and its dynamics is compared with ordinary numerical solutions. A high similarity is observed for visual images of attractors, power spectra, bifurcation diagrams and Lyapunov exponents.
We study the hyperchaos formation scenario in the modified Anishchenko–Astakhov generator. The scenario is connected with the existence of sequence of secondary torus bifurcations of resonant cycles preceding the hyperchaos emergence. This bifurcation cascade leads to the birth of the hierarchy of saddle-focus cycles with a two-dimensional unstable manifold as well as of saddle hyperchaotic sets resulting from the period-doubling cascades of unstable resonant cycles. Hyperchaos is born as a result of an inverse cascade of bifurcations of the emergence of discrete spiral Shilnikov attractors, accompanied by absorbing the cycles constituting this hierarchy.
Appearance of chaotic dynamics as a result of multi-frequency tori destruction is carried out on the example of a model of a multimode generator. Quasiperiodic bifurcations occurring with multi-frequency tori are discussed in the context of the Landau-Hopf scenario. Structure of the parameter space is studied, areas with various chaotic dynamics, including chaos and hyperchaos, are revealed. Scenarios of the development of chaotic dynamics are described, the features of chaotic signals of various types are revealed.
In this paper, we consider a class of orientation-preserving Morse–Smale diffeomorphisms defined on an orientable surface. The papers by Bezdenezhnykh and Grines showed that such diffeomorphisms have a finite number of heteroclinic orbits. In addition, the classification problem for such diffeomorphisms is reduced to the problem of distinguishing orientable graphs with substitutions describing the geometry of a heteroclinic intersection. However, such graphs generally do not admit polynomial discriminating algorithms. This article proposes a new approach to the classification of these cascades. For this, each diffeomorphism under consideration is associated with a graph that allows the construction of an effective algorithm for determining whether graphs are isomorphic. We also identified a class of admissible graphs, each isomorphism class of which can be realized by a diffeomorphism of a surface with an orientable heteroclinic. The results obtained are directly related to the realization problem of homotopy classes of homeomorphisms on closed orientable surfaces. In particular, they give an approach to constructing a representative in each homotopy class of homeomorphisms of algebraically finite type according to the Nielsen classification, which is an open problem today.
The class of quasi-chain graphs is an extension of the well-studied class of chain graphs. The latter class enjoys many nice and important properties, such as bounded clique-width, implicit representation, well-quasi-ordering by induced subgraphs, etc. The class of quasi-chain graphs is substantially more complex. In particular, this class is not well-quasi-ordered by induced subgraphs, and the clique-width is not bounded in it. In the present paper, we show that the universe of quasichain graphs is at least as complex as the universe of permutations by establishing a bijection between the class of all permutations and a subclass of quasi-chain graphs. This implies, in particular, that the induced subgraph isomorphism problem is NP-complete for quasi-chain graphs. On the other hand, we propose a decomposition theorem for quasi-chain graphs that implies an implicit representation for graphs in this class and efficient solutions for some algorithmic problems that are generally intractable.
In this paper, we consider regular topological flows on closed n-manifolds. Such flows have a hyperbolic (in the topological sense) chain recurrent set consisting of a finite number of fixed points and periodic orbits. The class of such flows includes, for example, Morse – Smale flows, which are closely related to the topology of the supporting manifold. This connection is provided by the existence of the Morse –Bott energy function for the Morse – Smale flows. It is well known that, starting from dimension 4, there exist nonsmoothing topological manifolds, on which dynamical systems can be considered only in a continuous category. The existence of continuous analogs of regular flows on any topological manifolds is an open question, as is the existence of energy functions for such flows. In this paper, we study the dynamics of regular topological flows, investigate the topology of the embedding and the asymptotic behavior of invariant manifolds of fixed points and periodic orbits. The main result is the construction of the Morse –Bott energy function for such flows, which ensures their close connection with the topology of the ambient manifold.
In this article we study the plasma motion in the transitional layer of a coronal loop randomly driven at one of its footpoints in the thin-tube and thin-boundary-layer (TTTB) approximation. We introduce the average of the square of a random function with respect to time. This average can be considered as the square of the oscillation amplitude of this quantity. Then we calculate the oscillation amplitudes of the radial and azimuthal plasma displacement as well as the perturbation of the magnetic pressure. We find that the amplitudes of the plasma radial displacement and the magnetic-pressure perturbation do not change across the transitional layer. The amplitude of the plasma radial displacement is of the same order as the driver amplitude. The amplitude of the magnetic-pressure perturbation is of the order of the driver amplitude times the ratio of the loop radius to the loop length squared. The amplitude of the plasma azimuthal displacement is of the order of the driver amplitude times Re1/6, where Re is the Reynolds number. It has a peak at the position in the transitional layer where the local Alfvén frequency coincides with the fundamental frequency of the loop kink oscillation. The ratio of the amplitude near this position and far from it is of the order of , where is the ratio of thickness of the transitional layer to the loop radius. We calculate the dependence of the plasma azimuthal displacement on the radial distance in the transitional layer in a particular case where the density profile in this layer is linear
The study of deformations of Lie algebras is related to the problem of classification of simple Lie algebras over fields of small characteristics. The classification of finite-dimensional simple Lie algebras over algebraically closed fields of characteristic p > 3 is completed. Over fields of characteristic 2, a large number of examples of Lie algebras are constructed that do not fit into previously known schemes. Description of deformations of classical Lie algebras gives new examples of simple Lie algebras and gives a possibility to describe known examples as deformations of classical Lie algebras. In this paper, we describe global deformations of Lie algebras of the type Dn and their quotient algebras Dn by the center in the case of a field of characteristic 2.
The weighted vertex coloring problem for a given weighted graph is to minimize the number of colors so that for each vertex the number of the colors that are assigned to this vertex is equal to its weight and the assigned sets of vertices are disjoint for any adjacent vertices. For all but four hereditary classes that are defined by two connected 5-vertex induced prohibitions, the computational complexity is known of the weighted vertex coloring problem with unit weights. For four of the six pairwise intersections of these four classes, the solvability was proved earlier of the weighted vertex coloring problem in time polynomial in the sum of the vertex weights. Here we justify this fact for the remaining two intersections.
The article is considering the problem of increasing the performance and accuracy of video face identification. We examine the selection of the several best video frames using various techniques for assessing the quality of images. In contrast to traditional methods with estimation of image brightness/contrast, we propose to utilize the deep learning techniques that estimate the frame quality by using the lightweight convolutional neural network. In order to increase the effectiveness of the frame quality assessment step, we propose to distill knowledge of the cumbersome existing FaceQNet model for which there is no publicly available training dataset. The selected K-best frames are used to describe an input set of frames with a single average descriptor suitable for the nearest neighbor classifier. The proposed algorithm is compared with the traditional face feature extraction for each frame, as well as with the known clustering methods for a set of video frames.
By combining the analytical approach and the test particle simulation, we examine the possibility of strong amplification of appropriate weak electromagnetic pulse in a magnetized plasma layer. Amplification occurs with the rate of instability of the hydrodynamic type through the beam plasma amplifier mechanism (BPA). This mechanism provide the excitation of intense rapidly changing radiation at an extraordinary plasma mode without a significant anisotropy of the distribution function. In a plasma with an nearly stable distribution of particles in the velocity space, the gain of radiation is determined by the plasma thermal to magnetic pressure ratio (β). Our simulations indicate the importance of the beam plasma amplifier mechanism for the interactions of active particle cloud with electromagnetic waves. Actually, a fraction of plasma shot noise can sharply amplify and turn into intense rapidly changing discrete emissions. Discrete emissions can be excited both in laboratory plasma installations and in the form of VLF chorus in the Earth's magnetosphere.
One of the ways to join the connectionist approach and the symbolic paradigm is Tensor Product Variable Binding. It was initially devoted to building distributed representation of recursive structures for neural networks to use it as the input. Structures are an essential part of both formal and natural languages and appear in syntactic trees, grammar, semantic interpretation. A human mind smoothly operates with the appearing problems on the neural level, and it is naturally scalable and robust. The question arises of whether it is possible to translate traditional symbolic algorithms to the sub-symbolic level to reuse performance and computational gain of the neural networks for general tasks. However, several aspects of Tensor Product Variable Binding lack attention in public research, especially in building such a neural architecture that performs computations according to the mathematical model without preliminary training. In this paper, those implementation aspects are addressed. A proposed novel design for the decoding network translates a tensor to a corresponding recursive structure with the arbitrary level of nesting. Also, several complex topics about encoding such structures in the distributed representation or tensor are addressed. Both encoding and decoding neural networks are built with the Keras framework’s help and are analyzed from the perspective of applied value. The proposed design continues the series of papers dedicated to building a robust bridge between two computational paradigms: connectionist and symbolic.
This paper deals with one-dimensional factor maps for the geometric model of Lorenz-type attractors in the form of two-parameter family of Lorenz maps on the interval 𝐼=[−1,1]I=[−1,1] given by 𝑇𝑐,𝜈(𝑥)=(−1+𝑐⋅|𝑥|𝜈)⋅𝑠𝑖𝑔𝑛(𝑥)Tc,ν(x)=(−1+c⋅|x|ν)⋅sign(x). This is the normal form for splitting the homoclinic loop with additional degeneracy in flows with symmetry that have a saddle equilibrium with a one-dimensional unstable manifold. Due to L. P. Shilnikov’ results, such a bifurcation (under certain conditions) corresponds to the birth of the Lorenz attractor. We indicate those regions in the parameter plane where the topological entropy depends monotonically on the parameter 𝑐c, as well as those for which the monotonicity does not take place. Also, we indicate the corresponding bifurcations for the Lorenz attractors.