Distributed Kernel Regression: An Algorithm for Training Collaboratively

摘要

This paper addresses the problem of distributed learning under communicationconstraints, motivated by distributed signal processing in wireless sensornetworks and data mining with distributed databases. After formalizing ageneral model for distributed learning, an algorithm for collaborativelytraining regularized kernel least-squares regression estimators is derived.Noting that the algorithm can be viewed as an application of successiveorthogonal projection algorithms, its convergence properties are investigatedand the statistical behavior of the estimator is discussed in a simplifiedtheoretical setting.