Distributed on-line anomaly detection using kernel methods

In this paper we develop an online distributed learning algorithm for unlabeled data. Here a goal is to detect anomalies from streaming data. The assumption is that data is drawn from a probability distribution, but that at some unknown time data could be drawn from a different distribution. We assume the data comes from distributed sources and develop an online Federated Learning (FL) algorithm based on online kernel methods using Random Fourier Features (RFF). This work combines previous work on online unsupervised kernel algorithms with online supervised kernel algorithms using FL and RFF. There are many applications of this work including detecting bad power grid data.