All issues

Finding the global minimum of a nonconvex function is one of the key and most difficult problems of the modern optimization. In this paper we consider special classes of nonconvex problems which have a clear and distinct global minimum.

In the first part of the paper we consider two classes of «good» nonconvex functions, which can be bounded below and above by a parabolic function. This class of problems has not been widely studied in the literature, although it is rather interesting from an applied point of view. Moreover, for such problems first-order and higher-order methods may be completely ineffective in finding a global minimum. This is due to the fact that the function may oscillate heavily or may be very noisy. Therefore, our new methods use only zero-order information and are based on grid search. The size and fineness of this grid, and hence the guarantee of convergence speed and oracle complexity, depend on the «goodness» of the problem. In particular, we show that if the function is bounded by fairly close parabolic functions, then the complexity is independent of the dimension of the problem. We show that our new methods converge with a linear convergence rate $\log(1/\varepsilon)$ to a global minimum on the cube.

In the second part of the paper, we consider the nonconvex optimization problem from a different angle. We assume that the target minimizing function is the sum of the convex quadratic problem and a nonconvex «noise» function proportional to the distance to the global solution. Considering functions with such noise assumptions for zero-order methods is new in the literature. For such a problem, we use the classical gradient-free approach with gradient approximation through finite differences. We show how the convergence analysis for our problems can be reduced to the standard analysis for convex optimization problems. In particular, we achieve a linear convergence rate for such problems as well.

Experimental results confirm the efficiency and practical applicability of all the obtained methods.

Gradient-free optimization methods or zeroth-order methods are widely used in training neural networks, reinforcement learning, as well as in industrial tasks where only the values of a function at a point are available (working with non-analytical functions). In particular, the method of error back propagation in PyTorch works exactly on this principle. There is a well-known fact that computer calculations use heuristics of floating-point numbers, and because of this, the problem of finiteness of the mantissa arises.

In this paper, firstly, we reviewed the most popular methods of gradient approximation: Finite forward/central difference (FFD/FCD), Forward/Central wise component (FWC/CWC), Forward/Central randomization on $l_2$ sphere (FSSG2/CFFG2); secondly, we described current theoretical representations of the noise introduced by the inaccuracy of calculating the function at a point: adversarial noise, random noise; thirdly, we conducted a series of experiments on frequently encountered classes of problems, such as quadratic problem, logistic regression, SVM, to try to determine whether the real nature of machine noise corresponds to the existing theory. It turned out that in reality (at least for those classes of problems that were considered in this paper), machine noise turned out to be something between adversarial noise and random, and therefore the current theory about the influence of the mantissa limb on the search for the optimum in gradient-free optimization problems requires some adjustment.

The paper is devoted to convex-concave saddle point problems where the objective is a sum of a large number of functions. Such problems attract considerable attention of the mathematical community due to the variety of applications in machine learning, including adversarial learning, adversarial attacks and robust reinforcement learning, to name a few. The individual functions in the sum usually represent losses related to examples from a data set. Additionally, the formulation admits a possibly nonsmooth composite term. Such terms often reflect regularization in machine learning problems. We assume that the dimension of one of the variable groups is relatively small (about a hundred or less), and the other one is large. This case arises, for example, when one considers the dual formulation for a minimization problem with a moderate number of constraints. The proposed approach is based on using Vaidya’s cutting plane method to minimize with respect to the outer block of variables. This optimization algorithm is especially effective when the dimension of the problem is not very large. An inexact oracle for Vaidya’s method is calculated via an approximate solution of the inner maximization problem, which is solved by the accelerated variance reduced algorithm Katyusha. Thus, we leverage the structure of the problem to achieve fast convergence. Separate complexity bounds for gradients of different components with respect to different variables are obtained in the study. The proposed approach is imposing very mild assumptions about the objective. In particular, neither strong convexity nor smoothness is required with respect to the low-dimensional variable group. The number of steps of the proposed algorithm as well as the arithmetic complexity of each step explicitly depend on the dimensionality of the outer variable, hence the assumption that it is relatively small.

This paper presents the results of experimental validation of some structural issues concerning the practical use of methods to overcome catastrophic forgetting of neural networks. A comparison of current effective methods like EWC (Elastic Weight Consolidation) and WVA (Weight Velocity Attenuation) is made and their advantages and disadvantages are considered. It is shown that EWC is better for tasks where full retention of learned skills is required on all the tasks in the training queue, while WVA is more suitable for sequential tasks with very limited computational resources, or when reuse of representations and acceleration of learning from task to task is required rather than exact retention of the skills. The attenuation of the WVA method must be applied to the optimization step, i. e. to the increments of neural network weights, rather than to the loss function gradient itself, and this is true for any gradient optimization method except the simplest stochastic gradient descent (SGD). The choice of the optimal weights attenuation function between the hyperbolic function and the exponent is considered. It is shown that hyperbolic attenuation is preferable because, despite comparable quality at optimal values of the hyperparameter of the WVA method, it is more robust to hyperparameter deviations from the optimal value (this hyperparameter in the WVA method provides a balance between preservation of old skills and learning a new skill). Empirical observations are presented that support the hypothesis that the optimal value of this hyperparameter does not depend on the number of tasks in the sequential learning queue. And, consequently, this hyperparameter can be picked up on a small number of tasks and used on longer sequences.

A model of combustion of premixed mixture of gases with one global chemical reaction is considered, the model includes equations of the second order for temperature of mixture and concentrations of fuel and oxidizer, and the right-hand sides of these equations contain the reaction rate function. This function depends on five unknown parameters of the global reaction and serves as approximation to multistep reaction mechanism. The model is reduced, after replacement of variables, to one equation of the second order for temperature of mixture that transforms to a first-order equation for temperature derivative depending on temperature that contains a parameter of flame propagation velocity. Thus, for computing the parameter of burning velocity, one has to solve Dirichlet problem for first-order equation, and after that a model dependence of burning velocity on mixture equivalence ratio at specified reaction rate parameters will be obtained. Given the experimental data of dependence of burning velocity on mixture equivalence ratio, the problem of optimal selection of reaction rate parameters is stated, based on minimization of the mean square deviation of model values of burning velocity on experimental ones. The aim of our study is analysis of uniqueness of this problem solution. To this end, we apply computational experiment during which the problem of global search of optima is solved using multistart of gradient descent. The computational experiment clarifies that the inverse problem in this statement is underdetermined, and every time, when running gradient descent from a selected starting point, it converges to a new limit point. The structure of the set of limit points in the five-dimensional space is analyzed, and it is shown that this set can be described with three linear equations. Therefore, it might be incorrect to tabulate all five parameters of reaction rate based on just one match criterion between model and experimental data of flame propagation velocity. The conclusion of our study is that in order to tabulate reaction rate parameters correctly, it is necessary to specify the values of two of them, based on additional optimality criteria.

The paper investigates a potential solution to the problem of Self-Interference Cancellation (SIC) encountered in the design of In-Band Full-Duplex (IBFD) communication systems. The suppression of selfinterference is implemented in the digital domain using multilayer nonlinear models adapted via the gradient descent method. The presence of local optima and saddle points in the adaptation of multilayer models prevents the use of second-order methods due to the indefinite nature of the Hessian matrix.

This work proposes the use of the Mixed Newton Method (MNM), which incorporates information about the second-order mixed partial derivatives of the loss function, thereby enabling a faster convergence rate compared to traditional first-order methods. By constructing the Hessian matrix solely with mixed second-order partial derivatives, this approach mitigates the issue of “getting stuck” at saddle points when applying the Mixed Newton Method for adapting multilayer nonlinear self-interference compensators in full-duplex system design.

The Hammerstein model with complex parameters has been selected to represent nonlinear selfinterference. This choice is motivated by the model’s ability to accurately describe the underlying physical properties of self-interference formation. Due to the holomorphic property of the model output, the Mixed Newton Method provides a “repulsion” effect from saddle points in the loss landscape.

The paper presents convergence curves for the adaptation of the Hammerstein model using both the Mixed Newton Method and conventional gradient descent-based approaches. Additionally, it provides a derivation of the proposed method along with an assessment of its computational complexity.

The paper reviews the existing approaches to calculating the destruction of solids. The main attention is paid to algorithms using a unified approach to the calculation of deformation both for nondestructive and for the destroyed states of the material. The thermodynamic derivation of the unified rheological relationships taking into account the elastic, viscous and plastic properties of materials and describing the loss of the deformation resistance ability with the accumulation of microdamages is presented. It is shown that the mathematical model under consideration provides a continuous dependence of the solution on input parameters (parameters of the material medium, initial and boundary conditions, discretization parameters) with softening of the material.

Explicit and implicit non-matrix algorithms for calculating the evolution of deformation and fracture development are presented. Non-explicit schemes are implemented using iterations of the conjugate gradient method, with the calculation of each iteration exactly coinciding with the calculation of the time step for two-layer explicit schemes. So, the solution algorithms are very simple.

The results of solving typical problems of destruction of solid deformable bodies for slow (quasistatic) and fast (dynamic) deformation processes are presented. Based on the experience of calculations, recommendations are given for modeling the processes of destruction and ensuring the reliability of numerical solutions.

The purpose of this work is to develop a universal computer program (solver) which solves kinetic Boltzmann equation for simulations of rarefied gas flows in complexly shaped devices. The structure of the solver is described in details. Its efficiency is demonstrated on an example of calculations of a modern many tubes Knudsen pump. The kinetic Boltzmann equation is solved by finite-difference method on discrete grid in spatial and velocity spaces. The differential advection operator is approximated by finite difference method. The calculation of the collision integral is based on the conservative projection method.

In the developed computational program the unstructured spatial mesh is generated using GMSH and may include prisms, tetrahedrons, hexahedrons and pyramids. The mesh is denser in areas of flow with large gradients of gas parameters. A three-dimensional velocity grid consists of cubic cells of equal volume.

A huge amount of calculations requires effective parallelization of the algorithm which is implemented in the program with the use of Message Passing Interface (MPI) technology. An information transfer from one node to another is implemented as a kind of boundary condition. As a result, every MPI node contains the information about only its part of the grid.

The main result of the work is presented in the graph of pressure difference in 2 reservoirs connected by a multitube Knudsen pump from Knudsen number. This characteristic of the Knudsen pump obtained by numerical methods shows the quality of the pump. Distributions of pressure, temperature and gas concentration in a steady state inside the pump and the reservoirs are presented as well.

The correctness of the solver is checked using two special test solutions of more simple boundary problems — test with temperature distribution between 2 planes with different temperatures and test with conservation of total gas mass.

The correctness of the obtained data for multitube Knudsen pump is checked using denser spatial and velocity grids, using more collisions in collision integral per time step.

Inverse geophysical problems are difficult to solve due to their mathematically incorrect formulation and large computational complexity. Geophysical exploration in frontier areas is even more complicated due to the lack of reliable geological information. In this case, inversion methods that allow interpretation of several types of geophysical data together are recognized to be of major importance. This paper is dedicated to one of such inversion methods, which is based on minimization of the determinant of the Gram matrix for a set of model vectors. Within the framework of this approach, we minimize a nonlinear functional, which consists of squared norms of data residual of different types, the sum of stabilizing functionals and a term that measures the structural similarity between different model vectors. We apply this approach to seismic and electromagnetic synthetic data set. Specifically, we study joint inversion of acoustic pressure response together with controlled-source electrical field imposing structural constraints on resulting electrical conductivity and P-wave velocity distributions.

We start off this note with the problem formulation and present the numerical method for inverse problem. We implemented the conjugate-gradient algorithm for non-linear optimization. The efficiency of our approach is demonstrated in numerical experiments, in which the true 3D electrical conductivity model was assumed to be known, but the velocity model was constructed during inversion of seismic data. The true velocity model was based on a simplified geology structure of a marine prospect. Synthetic seismic data was used as an input for our minimization algorithm. The resulting velocity model not only fit to the data but also has structural similarity with the given conductivity model. Our tests have shown that optimally chosen weight of the Gramian term may improve resolution of the final models considerably.

The work is devoted to the construction of efficient and applicable to real tasks first-order methods of convex optimization, that is, using only values of the target function and its derivatives. Construction uses OGMG, fast gradient method which is optimal by complexity, but requires to know the Lipschitz constant for gradient and the strong convexity constant to determine the number of steps and step length. This requirement makes practical usage very hard. An adaptive on the constant for strong convexity algorithm ACGM is proposed, based on restarts of the OGM-G with update of the strong convexity constant estimate, and an adaptive on the Lipschitz constant for gradient ALGM, in which the use of OGM-G restarts is supplemented by the selection of the Lipschitz constant with verification of the smoothness conditions used in the universal gradient descent method. This eliminates the disadvantages of the original method associated with the need to know these constants, which makes practical usage possible. Optimality of estimates for the complexity of the constructed algorithms is proved. To verify the results obtained, experiments on model functions and real tasks from machine learning are carried out.