Causal Discovery for Linear Models with Latent Confounding via Multi-domain Identifiable Representation Learning
In contrast, we propose Linear ADMG Learning (LinADMG), a framework for learning ADMGs for linear causal models through identifiable representation learning.
ProofPaper ↗
Key points
- Causal discovery with latent confounding is particularly challenging, but a common setting in real-world data.
- In this work, we focus on linear causal models with latent confounding, represented by acyclic directed mixed graphs (ADMGs).
- We show that the identifiability of the noise variables is sufficient for recovering all direct causal relations, and that, under an additional latent-confounding no-cancellation condition, LinADMG can identify the whole ADMG.
- We implement LinADMG with four identifiable representation learning approaches from the literature and validate the methods on simulated and semisynthetic datasets.
Sources (1)
- [1]Causal Discovery for Linear Models with Latent Confounding via Multi-domain Identifiable Representation LearningarXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 20, 12:43 PM
In contrast, we propose Linear ADMG Learning (LinADMG), a framework for learning ADMGs for linear causal models through identifiable representation learning.
Causal discovery with latent confounding is particularly challenging, but a common setting in real-world data.
Extractive summary: sentences quoted from the sources.