Unmeasured Confounding and More Recent Developments/Challenges in Causal Discovery

on YouTube ↗01h 08m

" summary

Covers practical differences between MAGs/PAGs and ADMGs: an edge flagged as visible edges implies no unmeasured confounder between its endpoints and can be read algorithmically from a PAG; no general GES-style greedy search exists for MAGs/PAGs (GFCI combines GES then FCI for speed but not full guarantees). Shows a parametric path: ICA-LiNGAM recovers a unique DAG under linear non‑Gaussian errors using ICA, while NoTears casts DAG search as a continuous constrained optimization (trace exp acyclicity) — methods work but need strong assumptions and struggle at large p.

AI-written summary · about

" chapters (15)

1. 0:00Introduction
2. 1:35Visible Edges
3. 8:30Scorebased Search
4. 15:05Alternative Approaches
5. 17:33Nonparametric Structural Equation
6. 23:50LIGAM
7. 25:13Nongaussianity
8. 34:21Latent Variables
9. 38:23The General
10. 39:33Recent Work
11. 44:23Another Question
12. 50:13Post nonlinear causal model
13. 51:51A Different Paradigm
14. 54:24NoTiers
15. 58:30Scores

" related talks

• 31mIntroduction to Causal Inference using pgmpyAnkur Ankan · PyData Amsterdam 2024
• 32mNew Approaches To Learning Nonparametric (Latent) Causal Graphical ModelsBryon Aragam · Algorithmic Aspects of Causal Inference 2022
• 31mCollaborative Causal Discovery With Atomic InterventionsShiva Kasiviswanathan · Algorithmic Aspects of Causal Inference 2022
• 01h 01mML for discrete optimization: Theoretical foundationsEllen Vitercik · Graph Learning Meets Theoretical Computer Science 2025
• 41mOn the Path to Causal InferenceMark Farragher · PyData London 2019
• 23mAgentic GraphRAG with MAGMA Manifold Agency with Graph Models and AugmentationAndreas Kollegger · Devworld Conference 2025
• 28mLearning Causal Representations From Unknown InterventionsKun Zhang · Learning from Interventions 2022
• 32mLearning Ill-Conditioned Gaussian Graphical ModelsRaghu Meka · Algorithmic Aspects of Causal Inference 2022