Large Scale Graph Representation Learning

One assumption that is commonly used in machine learning is that samples are statistically independent. In effect, each sample of data doesn’t tell you anything about any other sample in the dataset. This is not true for all types of data; there are some types of datasets where relationships between samples can be modeled as a graph. One example is the yelp dataset, where users and businesses can be represented by nodes, and user reviews can be represented by edges. Graph neural networks(GNNs) are a type of machine learning architecture that take advantage of the extra information given by graphs. However, they are much slower than traditional neural networks and they are currently designed for static, unchanging graphs. The objective of this research is to address both issues: to develop methods to increase the training speed of GNNs on large graphs and to adapt the architecture for time-varying information. Improvements the GNN methods will also yield improvements to downstream tasks such as recommending you a new restaurant based on your previous reviews.

Faculty Supervisor:

Sushant Sachdeva

Student:

Partner:

Microsoft Canada

Discipline:

Computer science

Sector:

Information and cultural industries

University:

University of Toronto

Program: