py-why
[FR] Weakly supervised causal representation learning
Perhaps a bit non-standard for this library, but would causal learn be interested in providing some causal representation learning algorithms which work well on low-level data (pixels, etc)?
Maybe as a start, a weakly supervised CRL algorithm, like the one presented in Brehmer et al (2022)?
I would be interested in contributing if so!
Reference: Brehmer, J., De Haan, P., Lippe, P., and Cohen, T. Weakly supervised causal representation learning. arXiv preprint arXiv:2203.16437, 2022.
Yeah, definitely this is something on the plan. But as you mentioned, it is a little bit different from the current goal of the library, which mainly focuses on discovering the structure (latent or observed) among variables. As a result, different infrastructures might be needed to support a wide range of CRL works. Would you mind ping me via email ([email protected]), so that I could let you know when we start to work on these CRL methods, and discuss about it if you are still interested?
