Skip to yearly menu bar Skip to main content


Weakly supervised causal representation learning

Johann Brehmer · Pim de Haan · Phillip Lippe · Taco Cohen

Hall J (level 1) #514

Keywords: [ Causality ] [ disentangled representation learning ] [ causal representation learning ] [ causal discovery ]


Learning high-level causal representations together with a causal model from unstructured low-level data such as pixels is impossible from observational data alone. We prove under mild assumptions that this representation is however identifiable in a weakly supervised setting. This involves a dataset with paired samples before and after random, unknown interventions, but no further labels. We then introduce implicit latent causal models, variational autoencoders that represent causal variables and causal structure without having to optimize an explicit discrete graph structure. On simple image data, including a novel dataset of simulated robotic manipulation, we demonstrate that such models can reliably identify the causal structure and disentangle causal variables.

Chat is not available.