Personalized Learning Using Kernel Methods and Random Fourier Features

Distributed learning algorithms have become an increasingly popular research topic due to the increasing proliferation of IoT devices and sensor networks with edge processing capabilities. This paper focuses on a distributed learning algorithm, Personalized Learning. In Personalized Learning, a population of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">N</tex> edge devices learns its similarities while preserving its learnings individually. Personalized Learning consists of a two-step iterative approach: <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">N</tex> edge processors are learning for each <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">L</tex> “canonical” model, and the characteristics of canonical models are estimated based on the learned edges. In earlier works, linear and logistic regression models were used to estimate characteristics in the first step. This work extends the previous work on the framework for personalized learning using non-linear data. This research improves the computational speed and performance of the current framework using Kernel Methods by Random Fourier Features for learning the first layer and compares these learning methods with linear models in simulation studies.