Design of Experiments for Learning from Online Networks

This PhD will develop novel methods to perform experiments on networks, combining the areas of Design of Experiments (DOE) and Network Science.

Focus will be on online experiments: consider the motivating example of optimally allocating adverts to social media users to maximise learning about relative strengths of the adverts. This is essentially A/B testing on a network, where an optimal strategy for deploying treatments is needed in real-time to maximise learning.

DOE is a statistical research area that allows scientists to maximise the amount of information gathered in an experiment: DOE helps to optimise the scientific method.

Network science is a growing academic discipline. Scientists from many areas analyse everything from transportation networks to protein interaction networks.

In networked experiments, the relationships between experimental units have a large effect on how we design experiments, and here we develop statistical theory and methods to optimise such experiments on large online networks.

There is some recent work on designing experiments on online experiments - and strong overlaps with "Active Learning" in the Machine Learning community. However, the general area of improving online experiments is relatively underdeveloped, at least with any academic rigour