All issues

The problem of online convex optimization naturally occurs in cases when there is an update of statistical information. The mirror descent method is well known for non-smooth optimization problems. Mirror descent is an extension of the subgradient method for solving non-smooth convex optimization problems in the case of a non-Euclidean distance. This paper is devoted to a stochastic variant of recently proposed Mirror Descent methods for convex online optimization problems with convex Lipschitz (generally, non-smooth) functional constraints. This means that we can still use the value of the functional constraint, but instead of (sub)gradient of the objective functional and the functional constraint, we use their stochastic (sub)gradients. More precisely, assume that on a closed subset of $n$-dimensional vector space, $N$ convex Lipschitz non-smooth functionals are given. The problem is to minimize the arithmetic mean of these functionals with a convex Lipschitz constraint. Two methods are proposed, for solving this problem, using stochastic (sub)gradients: adaptive method (does not require knowledge of Lipschitz constant neither for the objective functional, nor for the functional of constraint) and non-adaptivemethod (requires knowledge of Lipschitz constant for the objective functional and the functional of constraint). Note that it is allowed to calculate the stochastic (sub)gradient of each functional only once. In the case of non-negative regret, we find that the number of non-productive steps is $O$($N$), which indicates the optimality of the proposed methods. We consider an arbitrary proximal structure, which is essential for decisionmaking problems. The results of numerical experiments are presented, allowing to compare the work of adaptive and non-adaptive methods for some examples. It is shown that the adaptive method can significantly improve the number of the found solutions.

In this paper, we consider the conjugate gradient method for solving the problem of minimizing a quadratic function with additive noise in the gradient. Three concepts of noise were considered: antagonistic noise in the linear term, stochastic noise in the linear term and noise in the quadratic term, as well as combinations of the first and second with the last. It was experimentally obtained that error accumulation is absent for any of the considered concepts, which differs from the folklore opinion that, as in accelerated methods, error accumulation must take place. The paper gives motivation for why the error may not accumulate. The dependence of the solution error both on the magnitude (scale) of the noise and on the size of the solution using the conjugate gradient method was also experimentally investigated. Hypotheses about the dependence of the error in the solution on the noise scale and the size (2-norm) of the solution are proposed and tested for all the concepts considered. It turned out that the error in the solution (by function) linearly depends on the noise scale. The work contains graphs illustrating each individual study, as well as a detailed description of numerical experiments, which includes an account of the methods of noise of both the vector and the matrix.

In this paper, we consider conditional gradient methods for optimizing strongly convex functions. These are methods that use a linear minimization oracle, which, for a given vector $p \in \mathbb{R}^n$, computes the solution of the subproblem

\[ \text{Argmin}_{x\in X}{\langle p,\,x \rangle}. \]There are a variety of conditional gradient methods that have a linear convergence rate in a strongly convex case. However, in all these methods, the dimension of the problem is included in the rate of convergence, which in modern applications can be very large. In this paper, we prove that in the strongly convex case, the convergence rate of the conditional gradient methods in the best case depends on the dimension of the problem $ n $ as $ \widetilde {\Omega} \left(\!\sqrt {n}\right) $. Thus, the conditional gradient methods may turn out to be ineffective for solving strongly convex optimization problems of large dimensions.

Also, the application of conditional gradient methods to minimization problems of a quadratic form is considered. The effectiveness of the Frank – Wolfe method for solving the quadratic optimization problem in the convex case on a simplex (PageRank) has already been proved. This work shows that the use of conditional gradient methods to solve the minimization problem of a quadratic form in a strongly convex case is ineffective due to the presence of dimension in the convergence rate of these methods. Therefore, the Shrinking Conditional Gradient method is considered. Its difference from the conditional gradient methods is that it uses a modified linear minimization oracle. It's an oracle, which, for a given vector $p \in \mathbb{R}^n$, computes the solution of the subproblem \[ \text{Argmin}\{\langle p, \,x \rangle\colon x\in X, \;\|x-x_0^{}\| \leqslant R \}. \] The convergence rate of such an algorithm does not depend on dimension. Using the Shrinking Conditional Gradient method the complexity (the total number of arithmetic operations) of solving the minimization problem of quadratic form on a $ \infty $-ball is obtained. The resulting evaluation of the method is comparable to the complexity of the gradient method.

Multi-label classification models arise in various areas of life, which is explained by an increasing amount of information that requires prompt analysis. One of the mathematical methods for solving this problem is a plug-in approach, at the first stage of which, for each class, a certain ranking function is built, ordering all objects in some way, and at the second stage, the optimal thresholds are selected, the objects on one side of which are assigned to the current class, and on the other — to the other. Thresholds are chosen to maximize the target quality measure. The algorithms which properties are investigated in this article are devoted to the second stage of the plug-in approach which is the choice of the optimal threshold vector. This step becomes non-trivial if the $F$-measure of average precision and recall is used as the target quality assessment since it does not allow independent threshold optimization in each class. In problems of extreme multi-label classification, the number of classes can reach hundreds of thousands, so the original optimization problem is reduced to the problem of searching a fixed point of a specially introduced transformation $\boldsymbol V$, defined on a unit square on the plane of average precision $P$ and recall $R$. Using this transformation, two algorithms are proposed for optimization: the $F$-measure linearization method and the method of $\boldsymbol V$ domain analysis. The properties of algorithms are studied when applied to multi-label classification data sets of various sizes and origin, in particular, the dependence of the error on the number of classes, on the $F$-measure parameter, and on the internal parameters of methods under study. The peculiarity of both algorithms work when used for problems with the domain of $\boldsymbol V$, containing large linear boundaries, was found. In case when the optimal point is located in the vicinity of these boundaries, the errors of both methods do not decrease with an increase in the number of classes. In this case, the linearization method quite accurately determines the argument of the optimal point, while the method of $\boldsymbol V$ domain analysis — the polar radius.

A direct algorithm for solving a linear programming problem (LP), given in canonical form, is considered. The algorithm consists of two successive stages, in which the following LP problems are solved by a direct method: a non-degenerate auxiliary problem at the first stage and some problem equivalent to the original one at the second. The construction of the auxiliary problem is based on a multiplicative version of the Gaussian exclusion method, in the very structure of which there are possibilities: identification of incompatibility and linear dependence of constraints; identification of variables whose optimal values are obviously zero; the actual exclusion of direct variables and the reduction of the dimension of the space in which the solution of the original problem is determined. In the process of actual exclusion of variables, the algorithm generates a sequence of multipliers, the main rows of which form a matrix of constraints of the auxiliary problem, and the possibility of minimizing the filling of the main rows of multipliers is inherent in the very structure of direct methods. At the same time, there is no need to transfer information (basis, plan and optimal value of the objective function) to the second stage of the algorithm and apply one of the ways to eliminate looping to guarantee final convergence.

Two variants of the algorithm for solving the auxiliary problem in conjugate canonical form are presented. The first one is based on its solution by a direct algorithm in terms of the simplex method, and the second one is based on solving a problem dual to it by the simplex method. It is shown that both variants of the algorithm for the same initial data (inputs) generate the same sequence of points: the basic solution and the current dual solution of the vector of row estimates. Hence, it is concluded that the direct algorithm is an algorithm of the simplex method type. It is also shown that the comparison of numerical schemes leads to the conclusion that the direct algorithm allows to reduce, according to the cubic law, the number of arithmetic operations necessary to solve the auxiliary problem, compared with the simplex method. An estimate of the number of iterations is given.

Fluorescent imaging data are currently widely used in neuroscience and other fields. Genetically encoded sensors, based on fluorescent proteins, provide a wide inventory enabling scientiests to image virtually any process in a living cell and extracellular environment. However, especially due to the need for fast scanning, miniaturization, etc, the imaging data can be severly corrupred with multiplicative heteroscedactic noise, reflecting stochastic nature of photon emission and photomultiplier detectors. Deep learning architectures demonstrate outstanding performance in image segmentation and denoising, however they can require large clean datasets for training, and the actual data transformation is not evident from the network architecture and weight composition. On the other hand, some classical data transforms can provide for similar performance in combination with more clear insight in why and how it works. Here we propose an algorithm for denoising fluorescent dynamical imaging data, which is based on multilinear higher-order singular value decomposition (HOSVD) with optional truncation in rank along each axis and thresholding of the tensor of decomposition coefficients. In parallel, we propose a convenient paradigm for validation of the algorithm performance, based on simulated flurescent data, resulting from biophysical modeling of calcium dynamics in spatially resolved realistic 3D astrocyte templates. This paradigm is convenient in that it allows to vary noise level and its resemblance of the Gaussian noise and that it provides ground truth fluorescent signal that can be used to validate denoising algorithms. The proposed denoising method employs truncated HOSVD twice: first, narrow 3D patches, spanning the whole recording, are processed (local 3D-HOSVD stage), second, 4D groups of 3D patches are collaboratively processed (non-local, 4D-HOSVD stage). The effect of the first pass is twofold: first, a significant part of noise is removed at this stage, second, noise distribution is transformed to be more Gaussian-like due to linear combination of multiple samples in the singular vectors. The effect of the second stage is to further improve SNR. We perform parameter tuning of the second stage to find optimal parameter combination for denoising.

Semiclassical approximation formalism is developed for the multidimensional Fokker–Planck–Kolmogorov equation with non-local and nonlinear drift vector with respect to a small diffusion coefficient D, D→0, in the class of trajectory concentrated functions. The Einstein−Ehrenfest system of (0, M)-type is obtained. A family of semiclassical solutions localized around a point driven by the Einstein−Ehrenfest system accurate to O(D^(M+1)/2) is found.

The mathematical model of a three-layered Co/Cu/Co nanopillar for MRAM cell with one fixed and one free layer was investigated in the approximation of uniformly distributed magnetization. The anisotropy axis is perpendicular to the layers (so-called perpendicular anisotropy). Initially the magnetization of the free layer is oriented along the anisotropy axis in the position accepted to be “zero”. Simultaneous magnetic field and spinpolarized current engaging can reorient the magnetization to another position which in this context can be accepted as “one”. The mathematical description of the effect is based on the classical vector Landau–Lifshits equation with the dissipative term in the Gilbert form. In our model we took into account the interactions of the magnetization with an external magnetic field and such effective magnetic fields as an anisotropy and demagnetization ones. The influence of the spin-polarized injection current is taken into account in the form of Sloczewski–Berger term. The model was reduced to the set of three ordinary differential equations with the first integral. It was shown that at any current and field the dynamical system has two main equilibrium states on the axis coincident with anisotropy axis. It was ascertained that in contrast with the longitudinal-anisotropy model, in the model with perpendicular anisotropy there are no other equilibrium states. The stability analysis of the main equilibrium states was performed. The bifurcation diagrams characterizing the magnetization dynamics at different values of the control parameters were built. The classification of the phase portraits on the unit sphere was performed. The features of the dynamics at different values of the parameters were studied and the conditions of the magnetization reorientation were determined. The trajectories of magnetization switching were calculated numerically using the Runge–Kutta method. The parameter values at which limit cycles exist were determined. The threshold values for the switching current were found analytically. The threshold values for the structures with longitudinal and perpendicular anisotropy were compared. It was established that in the structure with the perpendicular anisotropy at zero field the switching current is an order lower than in the structure with the longitudinal one.

Equations of motion in Newtonian and Hamiltonian forms are used for classical molecular dynamics simulation of particle system time evolution. When Newton equations of motion are used for finding of particle coordinates and velocities in $N$-particle system it takes to solve $3N$ ordinary differential equations of second order at every time step. Traditionally numerical schemes of Verlet method are used for solving Newtonian equations of motion of molecular dynamics. A step of integration is necessary to decrease for Verlet numerical schemes steadiness conservation on sufficiently large time intervals. It leads to a significant increase of the volume of calculations. Numerical schemes of Verlet method with Hamiltonian conservation control (the energy of the system) at every time moment are used in the most software packages of molecular dynamics for numerical integration of equations of motion. It can be used two complement each other approaches to decrease of computational time in molecular dynamics calculations. The first of these approaches is based on enhancement and software optimization of existing software packages of molecular dynamics by using of vectorization, parallelization and special processor construction. The second one is based on the elaboration of efficient methods for numerical integration for equations of motion. A procedure for constructing of explicit, implicit and symmetric symplectic numerical schemes with given approximation accuracy in relation to integration step for solving of molecular dynamic equations of motion in Hamiltonian form is proposed in this work. The approach for construction of proposed in this work procedure is based on the following points: Hamiltonian formulation of equations of motion; usage of Taylor expansion of exact solution; usage of generating functions, for geometrical properties of exact solution conservation, in derivation of numerical schemes. Numerical experiments show that obtained in this work symmetric symplectic third-order accuracy scheme conserves basic properties of the exact solution in the approximate solution. It is more stable for approximation step and conserves Hamiltonian of the system with more accuracy at a large integration interval then second order Verlet numerical schemes.

Comparative analysis of two numerical methods for simulation of unsteady natural convection and thermal surface radiation within a differentially heated cubical cavity has been carried out. The considered domain of interest had two isothermal opposite vertical faces, while other walls are adiabatic. The walls surfaces were diffuse and gray, namely, their directional spectral emissivity and absorptance do not depend on direction or wavelength but can depend on surface temperature. For the reflected radiation we had two approaches such as: 1) the reflected radiation is diffuse, namely, an intensity of the reflected radiation in any point of the surface is uniform for all directions; 2) the reflected radiation is uniform for each surface of the considered enclosure. Mathematical models formulated both in primitive variables “velocity–pressure” and in transformed variables “vector potential functions – vorticity vector” have been performed numerically using finite volume method and finite difference methods, respectively. It should be noted that radiative heat transfer has been analyzed using the net-radiation method in Poljak approach.

Using primitive variables and finite volume method for the considered boundary-value problem we applied power-law for an approximation of convective terms and central differences for an approximation of diffusive terms. The difference motion and energy equations have been solved using iterative method of alternating directions. Definition of the pressure field associated with velocity field has been performed using SIMPLE procedure.

Using transformed variables and finite difference method for the considered boundary-value problem we applied monotonic Samarsky scheme for convective terms and central differences for diffusive terms. Parabolic equations have been solved using locally one-dimensional Samarsky scheme. Discretization of elliptic equations for vector potential functions has been conducted using symmetric approximation of the second-order derivatives. Obtained difference equation has been solved by successive over-relaxation method. Optimal value of the relaxation parameter has been found on the basis of computational experiments.

As a result we have found the similar distributions of velocity and temperature in the case of these two approaches for different values of Rayleigh number, that illustrates an operability of the used techniques. The efficiency of transformed variables with finite difference method for unsteady problems has been shown.