Compositional Generalization in Causal Inference

So can we start with this idea of, of compositional generalization in causal inference? Is this where you're trying to have that low dimensional vector representation be interpretable as high level variables. And then having say the directed acyclic graph pathways helps you like structure the compositional testing for such a system.Yeah, I think that's very good way of thinking about it. So yeah, I think it makes more sense to think about disentanglement as you're kind of, you're disentangling out the factors in this causal graph.