This academic paper addresses the concepts of identifiability and achievability in the context of intervention-based causal representation learning (CRL) under a general nonparametric latent causal model and an unknown transformation. The authors investigate both linear and general transformations, designing a score-based class of algorithms that provide both identifiability and achievability. The paper proves that one stochastic hard intervention per node is enough to guarantee identifiability for linear transformations, while two are needed for general transformations. The results are significant for the field of data science and machine learning.
Publication date: 1 Feb 2024
Project Page: https://arxiv.org/abs/2402.00849v1
Paper: https://arxiv.org/pdf/2402.00849
