This paper discusses causal representation learning in a nonparametric setting, focusing on multiple distributions without assuming hard interventions. The authors aim to develop solutions for this fundamental case, which can offer unique benefits compared to other assumptions such as parametric causal models. The study shows that under certain conditions, one can recover the moralized graph of the underlying directed acyclic graph. The recovered latent variables and their relations are related to the underlying causal model in a specific, nontrivial way. Experimental results support the theoretical claims made in the paper.
Publication date: 8 Feb 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2402.05052
