To build robust cause-effect inference tools, the causal sufficiency assumption often needs to be relaxed, as many real-world problems involve hidden common confounders (e.g., age or gender) that affect both variables. One proposed approach models these confounders by introducing correlation between the noise variables $N_X$ and $N_Y$ that affect variables $X$ and $Y$, or by modeling unobserved confounding effects using a new noise variable $N_{XY}$ within the generation process of $X$ and $Y$.