**Tobias Fritz** Max Planck Institute for Mathematics in the Sciences, Leipzig, Germany

**The Inflation Technique for Causal Inference**

**Watch video**

**Download video**

**Abstract:**

The problem of causal inference is to determine whether a given probability distribution on observed variables is compatible with some hypothetical Bayesian network structure. In the presence of hidden nodes (unobserved variables), this is a challenging problem for which no exact methods were previously known. The inflation technique provides such a method which is both useful in practice and has recently been shown to result in necessary and sufficient conditions. Finally, I will speculate on applications outside of causal inference, in particular on the problem of deciding whether a given function can be implemented

by a neural network of a given architecture.