9 July 2021 in Optimization Methods and Software
In this paper, we propose a general algorithmic framework for the first-order methods in optimization in a broad sense, including minimization problems, saddle-point problems and variational inequalities (VIs). This framework allows obtaining many known methods as a special case, the list including accelerated gradient method, composite optimization methods, level-set methods, Bregman proximal methods. The idea of the framework is based on constructing an inexact model of the main problem component, i.e. objective function in optimization or operator in VIs. Besides reproducing known results, our framework allows constructing new methods, which we illustrate by constructing a universal conditional gradient method and a universal method for VIs with a composite structure. This method works for smooth and non-smooth problems with optimal complexity without a priori knowledge of the problem's smoothness. As a particular case of our general framework, we introduce relative smoothness for operators and propose an algorithm for VIs with such operators. We also generalize our framework for relatively strongly convex objectives and strongly monotone VIs.
22 January 2021 in Journal of Inverse and Ill-posed Problems
We introduce primal and dual stochastic gradient oracle methods for decentralized convex optimization problems. Both for primal and dual oracles, the proposed methods are optimal in terms of the number of communication steps. However, for all classes of the objective, the optimality in terms of the number of oracle calls per node takes place only up to a logarithmic factor and the notion of smoothness. By using mini-batching technique, we show that the proposed methods with stochastic oracle can be additionally parallelized at each node. The considered algorithms can be applied to many data science problems and inverse problems.
1 November 2020 in Computational Mathematics and Mathematical Physics
Recently, it has been shown how, on the basis of the usual accelerated gradient method for solving problems of smooth convex optimization, accelerated methods for more complex problems (with a structure) and problems that are solved using various local information about the behavior of a function (stochastic gradient, Hessian, etc.) can be obtained. The term “accelerated methods” here means, on the one hand, the presence of some unified and fairly general way of acceleration. On the other hand, this also means the optimality of the methods, which can often be proved rigorously. In the present work, an attempt is made to construct in the same way a theory of accelerated methods for solving smooth convex-concave saddle-point problems with a structure. The main result of this article is the obtainment of in some sense necessary and sufficient conditions under which the complexity of solving nonlinear convex-concave saddle-point problems with a structure in the number of calculations of the gradients of composites in direct variables is equal in order of magnitude to the complexity of solving bilinear problems with a structure.