Causal Inference in Data Science

Kicking off a new MCDS series of talks - “data science across disciplines”.

At this first event in the series, the speakers at this hybrid event discussed causal inference in data science through the lens of their fields of expertise.

—

Talks:

Domain Adaptation as a Problem of Inference on Causal Graphical Models

Dr Mingming Gong

Causality has been playing an increasingly important role in various machine learning problems. In this talk, I will present our work on using causal models to build machine learning algorithms that could transfer to new data distributions. In particular, we cast the domain adaptation as a Bayesian inference problem on causal graphical models. The proposed method has shown promising results on image classification and wifi-localisation.

Applying Causal Inference to Spatial Data

A/Prof Martin Tomko and Mr Kamal Akbari

Causal inference, and causal inference methods are a quickly developing domain of data science. In this brief talk, we will illustrate why causal inference on data with a spatial component (where things are) is of high interest to a number of disciplines, but also why current causal inference methods are fundamentally inadequate for reasoning about causality in spatial data.

We will present some of our initial observations about spatial causal inference, and highlight aspects of the analytical workflows that deserve research interest.

Recording available soon