The paper investigates the impact of error in the set of Conditional Independencies (CIs) used to construct Probabilistic Graphical Models (PGMs) on the inferred CIs. It provides new insights into how these errors propagate in both directed (Bayesian Networks) and undirected (Markov Networks) PGMs. The study reveals that no guarantee can be provided for CIs inferred from the structure of undirected graphs, but such a guarantee exists for CIs inferred in directed graphical models. This makes the d-separation algorithm a sound and complete system for inferring approximate CIs.
Publication date: 21 Oct 2023
Project Page: https://arxiv.org/abs/2310.13942v1
Paper: https://arxiv.org/pdf/2310.13942
