combinatorial-generalization

This paper introduces relational structural causal models, extending structural causal models to settings with varying objects and relations. It provides theoretical results for identification and proposes relational neural causal models that outperform non-relational baselines on simulated traffic scenes.