Recent work has shown promising results in causal discovery by leveraging interventional data with gradient-based methods, even if the intervened variables are unknown. That previous work, however, assumes that the correspondence between samples and interventions is known, which is often unrealistic. We consider a scenario in which there is an underlying observational distribution that undergoes multiple interventions, but without knowledge of which intervention (if any) corresponds to each sample, and of how the interventions affect the system; i.e., the interventions are entirely latent. To address this scenario, we propose a method based on neural networks and variational inference, by framing the problem as that of learning a shared causal graph among an infinite mixture (under a Dirichlet process prior) of intervention structural causal models. Experiments with synthetic and real data show that our approach and its semi-supervised variant are able
to discover causal relations in this challenging scenario.