Machine Learning

1 readers

1 users here now

Community Rules:

Be nice. No offensive behavior, insults or attacks: we encourage a diverse community in which members feel safe and have a voice.
Make your post clear and comprehensive: posts that lack insight or effort will be removed. (ex: questions which are easily googled)
Beginner or career related questions go elsewhere. This community is focused in discussion of research and new projects that advance the state-of-the-art.
Limit self-promotion. Comments and posts should be first and foremost about topics of interest to ML observers and practitioners. Limited self-promotion is tolerated, but the sub is not here as merely a source for free advertisement. Such posts will be removed at the discretion of the mods.

founded 11 months ago

MODERATORS

communick@academy.garden

[D] Where does the 'ML' part of CAUSAL ML come in? (alien.top)

submitted 10 months ago by Ckdew@alien.top to c/machinelearning@academy.garden

5 comments fedilink hide all child comments

Looks like classic causal inference, the type would see in Pearl or Econometrics. I don't understand where the ML comes in.

you are viewing a single comment's thread
view the rest of the comments

[–] NeKleineMetAndalouse@alien.top 1 points 10 months ago

Learning causal effects from observational data is not straightforward, because of two reasons:

you have strong confounder bias: treatments are not assigned at random (unlike in an active experiment setup like an RCT). Patients who are more sick will receive more treatment. And
you never observe counterfactuals (if you treat a patient, you do not observe what would have happened if you had not treated this patient).

ML can help tackling the first issue by learning proper representations for your treated and non-treated groups. Simplified: if you manage to find representations for your control and treatment groups so that the distributions of your representations (which contain your confounding factors), you can make unbiased treatment effect estimates.

The second issue can be solved by using (semi-) synthetic data.