Hidden Variables in Linear Non-Gaussian Causal Models

Explore hidden variables in linear non-Gaussian causal models through this 32-minute conference talk by Elina Robeva from the University of British Columbia. Delve into the challenges of identifying causal relationships between random variables from observational data, particularly when hidden variables are present. Learn about linear structural equation models and their applications in causal structure learning. Discover how non-Gaussian variables in these models allow for full causal structure identification without interventions, contrasting with the limitations of Gaussian cases. Examine the use of high-order cumulant information to uncover the structure of linear non-Gaussian structural equation models with hidden variables. Understand the expanded approach that allows hidden variables to be common causes of multiple observed variables, moving beyond the traditional assumption of two observed variables per hidden variable. This talk, part of the Tensor Methods and Emerging Applications to the Physical and Data Sciences 2021 workshop, covers topics including structural equation causal models, nongaussian models, linear nongaussian causal models, hidden variables, multidirected edges, and cumulative tensors.

Intro
Structural Equation Causal Models
Nongaussian Structural Equation Models
Linear Nongaussian Causal Models
Example
Hidden variables
Multidirected edges
Cumulative tensor
Conclusion