Applications of The Inflation Technique for Causal Inference
Talkat Quantum Physics and Statistical Causal Models (Simons Institute workshop, 2022) · 2022-04-27..04-29
summary
Presents the inflation technique as a concrete method: build inflation graphs by taking IID copies of latent roots, convert expressible monomials and symmetry constraints into a linear program, and extract polynomial inequalities (e.g. from 1-2-1 or 2-2-2 inflations) that rule out incompatible triangle and related distributions (including Renou’s example). Shows that fan-out/cloning assumptions produce quantum-violable constraints, that a convergent hierarchy of relaxations recovers the true compatibility set as copies grow, and that do-conditional identifiability gives stronger, low-level inflations for multilayer DAGs.
AI-written summary · about