[AIRG] Causal discovery, generalizations of mutual information

[Aubrey Barnard <barnard@xxxxxxxxxxx>]

AIRG,

It seems like there was interest in learning more about discovering 
causality in observational [1] data. For those who are interested, I can 
suggest Judea Pearl's book on causality:

https://search.library.wisc.edu/catalog/9910129159702121
https://www.amazon.com/Causality-Reasoning-Inference-Judea-Pearl/dp/052189560X/

It treats causal discovery as Bayesian network structure learning. 
(Bayesian networks and other probabilistic graphical models are covered 
in most introductory machine learning books [2], although not all 
mention causality.)

The other main approach to causal discovery is observational studies. 
The primary difference is that structure learning jointly learns the 
structure of many variables whereas observational studies are limited to 
pairs of variables (and their covariates), but observational studies 
have better causal guarantees.

Also, Lucas said he would send me some resources on generalizations of 
mutual information. I will pass them on if he doesn't just send them to 
the whole list.

Causal discovery in observational data is the overarching topic of my 
research, so I would be happy to give a tutorial if there is interest. 
(Let me know personally.)

Aubrey


[1] As opposed to experimental / interventional data where one has 
control over some of the variables.

[2] Such as:
https://search.library.wisc.edu/catalog/9910317322702121
https://www.amazon.com/Machine-Learning-Probabilistic-Perspective-Computation/dp/0262018020/