Pearl absolutely uses graphical models, but he doesn’t use factor graphs as his main hammer. His whole causality empire is built on directed acyclic graphs (DAGs), not the bipartite factor-graph formalism from Koller’s side of the family.

Let me break this down without frying your neurons.

1. Does Judea Pearl use factor graphs?

Not really.

Pearl uses:

• Bayesian networks (directed PGMs)
• DAG-based structural causal models (SCMs)
• d-separation
• do-calculus
• backdoor/frontdoor criteria

All of this lives in the land of directed edges, not the “factor node vs variable node” bipartite setup.

Factor graphs are mostly used for:

• inference in undirected models (MRFs),
• message passing,
• sparse EBMs,
• turbo codes, LDPC decoding,