A1075
Title: Fast convergence of a federated expectation-maximization algorithm
Authors: Rajita Chandak - EPFL (Switzerland) [presenting]
Abstract: Data heterogeneity has been a long-standing bottleneck in studying the convergence rates of federated learning algorithms. The benefits of data-heterogeneity are illustrated through establishing convergence rates of the expectation-maximization (EM) algorithm for the federated mixture of k-linear regressions model (FMLR). The convergence rate of the EM algorithm is completely characterized under all regimes of m/n, where m is the number of nodes and n is the number of data points per node. Furthermore, theoretical and empirical implications of various standard assumptions are discussed in the literature, showcasing the need for careful consideration of statistical frameworks for federated algorithms.
