École Nationale Supérieure de
Techniques Avancées

Abstract

In machine learning, one of the big challenges is handling the sheer amount of data of the training sets. One way of processing huge datasets and turning them into more exploitable ones is through clustering. In this paper, we will focus on how clustering may improve the performance of first order stochastic gradient methods. We implement the state of the art techniques in clustering for stochastic gradient descent, and we seek to make improvements upon them. All of our code is available on our GitHub on https://github.com/yy214/cluster-for-opt.