12-3pm EDT, August 14, 2021 (Saturday)
Elena Zheleva (UIC) & David Arbour (Adobe Research)
The task of causal inference -- inferring the effect of interventions and counterfactuals from data -- is central to a vast number of scientific and industrial applications. To capture the noise, heterogeneity, and complex relationships in real-world data, it is customary to model data sources as relational systems and to reason about them probabilistically. Relations in data can be represented through heterogeneous networks in which nodes represent interdependent entities, such as people, companies, websites, and diseases, while edges denote different relationships between these entities, such as friendship, hyperlink, contribution, and spread of disease.
This tutorial will present state-of-the-art research on causal inference from network data, also known as causal inference with interference. We will start by motivating research in this area with real-world applications, such as measuring influence in social networks and market experimentation. We will discuss the challenges of applying existing causal inference techniques designed for independent and identically distributed (i.i.d.) data to relational data, some of the solutions that currently exist and the gaps and opportunities for future research. We will present existing network experiment designs for measuring different possible effects of interest. Then we will focus on causal inference from observational data, its representation, identification, and estimation. We will conclude with research on causal discovery in networks.
Stay tuned!