Academic

Abstract

Recovering latent structures is a key unsupervised learning task in network data, with applications spanning a multitude of disciplines. For example, identifying communities in webpages can lead to faster searches, classifying regions of the human brain network can be used to predict the onset of psychosis, and identifying communities of assets can help investors manage risk by investing in different communities of assets. However, the scale of these networks is massive, and most often it is impossible to obtain the full network information. This has necessitated the development of machine learning methods on networks with only a limited amount of available information. This talk will focus on recent advances in this direction in the context of clustering. In more detail, I will talk about some theoretical progress in the context of:

1. Spectral algorithms on networks with missing data, and
2. Graph representation learning algorithms such as DeepWalk/Node2vec.