
The problem of finding equilibrium in oligopolistic network problem is considered. The main attention is devoted to the taking into account the network line losses. A potential-like approach is suggested and results on a test example is given.
This paper presents a worst-case performance guarantee for a modification of the Brucker-Garey-Johnson algorithm applied to a problem on parallel machines, where a set of jobs is subject to in-tree precedence constraints and jobs have arbitrary processing times. The objective is to minimise maximum lateness. We show that this performance guarantee is tight for any maximum processing time of all jobs, that is assumed to be positive integer.
This paper compares several well-known algorithms for solving the traffic assignment problem and introduces a new method based on sequential quadratic programming (SQP). The proposed approach is designed with GPU-parallelization in mind, distinguishing it from traditional algorithms commonly used in this domain. Experimental results show that the method achieves performance comparable to existing techniques in a single-threaded environment while also demonstrating acceleration from GPU-parallelism. These findings highlight the advantages of the proposed algorithm for GPU-parallel computing environments while also identifying scenarios where classical shortest-path–based approaches remain more effective.
Cluster analysis is one of the central and most actively developed research directions in unsupervised machine learning. The Euclidean minimum-sum-of-squares clustering problem, referred to as the k-means problem, remains the most widely studied clustering model. A more general Bregman divergence-based clustering approach extends this classical model by replacing the squared Euclidean distance with a Bregman divergence for measuring the dissimilarity between data points and their nearest cluster centers. In the present work, we study this generalized clustering setting from the point of view of mathematical programming. In particular, we consider a practically important case of Bregman divergences, Kullback-Leibler divergence. We focus on the corresponding clustering problem as a nonconvex optimization problem and develop a local search method based on linearization of a nonconvex component in its DC representation. The developed approach is tested in a series of computational experiments and compared with conventional implementations of k-means and KL-means.
In mixed-integer linear programs a subset of variables may appear in the problem formulation only in the form of a certain linear combination, either in the original problem or after previous presolving steps eliminating variables and constraints. We present a multi-stage method to aggregate such combinations into a fewer number of variables.