Semiparametric approach to causal discovery

Structural equation models (SEM) are mathematical models that can be used to describe data generating processes. These links below would probably be helpful to overview the field of semiparametric methods including LiNGAM for estimating structural equation models. An important application of those methods is causal discovery. Non-Gaussianity and independence are the keys to model identification as in independent component analysis (ICA).

Here are tutorial slides at UAI2010 (and the references) and a survey paper (BHMK, 2014).

Quick links to topics:

• Basic linear models with no latent confounders (acyclic models, time series, cyclic models)  

• NEW Linear models with latent confounders and latent factors

• NEW Extensions (nonlinear models, discrete variables)

• NEW Related issues (causality and machine learning, testing, model fit and reliability, learning from multiple, benchmark datasets, others)

• NEW Applications and tailor-made methods

• NEW Related reviews and papers

• Software

---Updated on 9th April 2024