Communication-efficient solution of distributed variational inequalities using biased compression, data similarity and local updates

Variational inequalities constitute a broad class of problems with applications in a number of fields, including game theory, economics, and machine learning. Today’s practical applications of VIs are becoming increasingly computationally demanding. It is therefore necessary to employ distributed computations to solve such problems in a reasonable time. In this context, workers have to exchange data with each other, which creates a communication bottleneck. There are three main techniques to reduce the cost and the number of communications: the similarity of local operators, the compression of messages and the use of local steps on devices. There is an algorithm that uses all of these techniques to solve the VI problem and outperforms all previous methods in terms of communication complexity. However, this algorithm is limited to unbiased compression. Meanwhile, biased (contractive) compression leads to better results in practice, but it requires additional modifications within an algorithm and more effort to prove the convergence. In this work, we develop a new algorithm that solves distributed VI problems using data similarity, contractive compression and local steps on devices, derive the theoretical convergence of such an algorithm, and perform some experiments to show the applicability of the method.