Introduction to Causal Graphical Models: Graphs, d-separation, do-calculus

on YouTube ↗01h 06m

" summary

Defines d-separation precisely (chains, forks, colliders and descendant rules) and proves that d-separation implies conditional independence for all compatible distributions, plus the converse existence claim (if not d-separated there exists a compatible P showing dependence). Key tools: a lemma that nonadjacent nodes are d-separated by their parents, the tight active trail construction, and the Markov‑equivalence theorem: two DAGs share exactly the same d‑separations iff they have the same skeleton and immoralities. Also notes a positivity assumption for identifiability.

AI-written summary · about

" chapters (10)

1. 0:00Intro
2. 0:35Correct derivation
3. 7:02Example
4. 9:40Blocked trails
5. 13:18dseparation
6. 19:26z is a function
7. 23:16Theorem
8. 25:48Two graphs
9. 32:30Conditioning on parents
10. 43:34Tight active trails

" related talks

• 32mNew Approaches To Learning Nonparametric (Latent) Causal Graphical ModelsBryon Aragam · Algorithmic Aspects of Causal Inference 2022
• 44mMarginal DAG Models Equivalent to Another DAGRobin Evans · Quantum Physics and Statistical Causal Models (Simons Institute workshop, 2022)
• 48mThe Approximate Implication Problem For Probabilistic Graphical ModelsBatya Kenig · Algorithmic Aspects of Causal Inference 2022
• 01h 01mThe Mathematics of Causal Inference, with Reflections on Machine Learning and the Logic of ScienceJudea Pearl · Symposium on Visions of the Theory of Computing 2013
• 43mMarginal Determinism in Structured Decomposable CircuitsBenjie Wang · Probabilistic Circuits and Logic 2023
• 31mIntroduction to Causal Inference using pgmpyAnkur Ankan · PyData Amsterdam 2024
• 01h 00mLogic and Probabilistic Circuits 3YooJung Choi · Logic and Algorithms in Database Theory and AI Boot Camp
• 11mChoosing Between Dags Can Be HardRichard Guo · Meet the Fellows Welcome Event