Federated Learning with Variational Autoencoders

Hugo Dugdale

Keywords: machine learning, federated learning, variational autoencoders

TL;DR: Demonstrates feasibility of using federated learning to train a variational autoencoder

Abstract: In this work we investigate the feasibility of using federated learning to train a variational autoencoder capable of generated handwritten digits when trained on the MNIST dataset. It was found that using federated learning we were able to train a model that produced comparable results to a centralised model, both in image reconstructions and image generations.