From Structural Causal Models to Causal Abstraction Learning
Date:
In this talk we will discuss the formalization of two important aspects of human reasoning: causal reasoning and multi-level reasoning. First, we will introduce a common framework to model causal reasoning in machine learning, namely structural casual models (SCM). Next, we will discuss the problem of relating multiple causal models in order to exploit data and knowledge from multiple levels of abstraction or resolution. Finally, we will present the first recent attempts at learning relationships of abstractions between SCMs from data.