Collaborative Causal Discovery With Atomic Interventions
summary
Defines a collaborative model that recovers multiple causal graphs (as Maximal Ancestral Graphs (MAGs)) when entities’ MAGs satisfy an α–β clustering; algorithms sample selected nodes with atomic interventions to build partial MAGs, cluster entities, then load‑balance further interventions to finish recovery, achieving roughly logarithmic in M atomic interventions per entity in favorable regimes (stronger when β=0). Results include matching lower bounds (Ω(1/α) dependence), Monte‑Carlo high‑probability guarantees, and explicit dependence on MAG maximum degree Δ; proofs use randomized‑algorithm ideas.
AI-written summary · about
chapters (10)
- 0:00Intro
- 0:41Causal Discovery
- 1:19An Example
- 5:07Detour: How do we learn a Causal Graph?
- 8:37Meaning of the Edges of a MAG
- 9:46Maximal Ancestral Graphs (MAGs)
- 17:04Our Theoretical Guarantees
- 20:03Phase 1: Learning the Clustering
- 26:35A Word about the Lower Bound
- 28:33Experimental Evaluation
related talks
01h 02mClassical Strategies and Concepts in Causal Discovery
01h 08mUnmeasured Confounding and More Recent Developments/Challenges in Causal Discovery
23mAgentic GraphRAG with MAGMA Manifold Agency with Graph Models and Augmentation
01h 02mUnmeasured Confounding and More Recent Developments/Challenges in Causal Discovery
24mA Friendly Introduction To Causality
01h 06mIntroduction to Causal Graphical Models: Graphs, d-separation, do-calculus
28mLearning Causal Representations From Unknown Interventions
01h 01mIntroduction to Experimental Design