École Nationale Supérieure de
Techniques Avancées

Abstract

Many real world problems can be modeled as graphs. Usually, graph analysis is highly time and memory consuming. Graph representation learning reduces the dimension of the space in which we store the graph. We obtain representations of nodes in a lower dimensional space in which their relative positions are used to keep some desired properties of a given network. This allows for more efficient graph analysis. In this report, we define graph representation learning. Then, we present some state of the art methods in this research field. Afterwards, we upgrade the performance of these methods by applying some preprocessing to the graphs. The representation of the network is learnt by removing relatively unimportant edges so that we aim at obtaining representations more efficiently and enhance the performance of various downstream tasks (node classification and link prediction).