BGM-IV: Bayesian Generative Model for Nonlinear IV Regression
The innovative AI technique known as BGM-IV reinterprets nonlinear instrumental variable regression as posterior inference within a causally organized latent space. This method identifies latent factors related to confounding, outcomes, treatments, and nuisance variations, employing an IV-integrated pseudo-likelihood to address endogeneity. It is designed to focus on high-dimensional covariates and nonlinear structural effects.
Key facts
- BGM-IV is a latent Bayesian generative modeling approach
- It reframes nonlinear IV regression as posterior inference
- It infers latent components for shared confounding, outcome, treatment, and nuisance variation
- It uses an IV-integrated pseudo-likelihood to average over instruments
- The method addresses high-dimensional covariates and nonlinear structural effects
- The paper is published on arXiv with ID 2605.07029
- The announcement type is cross
Entities
Institutions
- arXiv