Decentralized Optimization
Overview
Decentralized optimization studies how networked agents can jointly minimize an objective or coordinate decisions while each uses local data and communicates with neighbors rather than relying on a central coordinator. Common approaches include gossip and consensus, decentralized stochastic gradient descent with gradient tracking, primal-dual and ADMM methods, and local updates paired with compressed or intermittent communication. These methods are tested in settings such as image and language-model training, online and bilevel learning, power systems, and multi-robot estimation. A central challenge is balancing communication cost, network topology, data heterogeneity, and robustness against noise or adversaries while retaining fast convergence and reliable solutions.
Latest papers
60 papers15 papers in the last four weeks, up 150% on the four weeks before. 0.1% of all new papers.