Causal learning and inference

Date:

Causal learning and discovery have received significant attention in the machine learning community. Causal learning allows us to answer the question about intervention (i.e., tinkering in the world) and imagination. This framework can mitigate the problem of robustness, adaptation, and explanation that is present in current machine learning practice.

In this workshop, I tried to introduce the basic ingredients of causal learning, do-calculus, the concept of d-separation and independence. In the end, I introduced some of the approaches that I find promising in recent research: such as finding invariant representation and the relationship
between invariant and causality, the multi-cause causal inference that arises in real scenarios like GWAS data analysis (whether multi-cause setting makes the problem identifiable or not).

The resources I have used for this workshop:

  1. Causality github page
  2. The Book of Why: The New Science of Cause and Effect
  3. Invariant Risk Minimization
  4. On Multi-Cause Causal Inference with Unobserved Confounding: Counterexamples, Impossibility, and Alternatives